CHANGELOG.md

Changelog

v0.68.6 - 2026-08-03

• Added one provider-neutral target-system blueprint kernel for software, workflows, services, agents, pipelines, and mixed systems. Whole-target claims now require a frozen provider registry and target snapshot, and every claimed capability binds its exact observed inputs and payloads.
• Bound each independently declared behavior owner to its exact code/workflow surfaces, input/state/output/effect/error/decision/order/retry/timeout/ completion contracts, source-independent semantics, resources, and accepted intent without promoting every discovered helper into a fake model.
• Added an independently re-audited project-test inventory with exact files, nodes, parameterization, assertions, source fingerprints, and bounded native checker artifacts; broad parent evidence can no longer fabricate child test coverage or certify stale source.
• Added content-addressed intent contributions from declared project inputs, including bounded Spark, OpenSpark, and changelog/history profiles, and bound accepted intent to the existing WorkContext and ModelRevisionSet lineage with explicit conflict, supersession, rejection, deferral, and unresolved-gap handling.
• Added exact parent/child interface mappings: every topology relation records which producer output becomes which consumer input, and disconnected, missing, or surface-mismatched model owners block static readiness.
• Added a real ModelTestAlignmentReport with its own fingerprint, model obligations, owner code contracts, checker designs, execution state, and visible gaps. A binding-report fingerprint can no longer impersonate it.
• Split checker design by behavior case and dimension so one assertion or native check cannot silently claim contradictory good/boundary/bad cases or all behavior dimensions. Parameter-case identity stays explicit and planned checks remain distinct from receipt-backed execution.
• Required project owners to declare their concrete cases and accepted checker designs. The generic builder no longer manufactures canonical cases or checker identities from observed source, while not_run remains an honest execution status for a complete static test design.
• Replaced FlowGuard's self-only blueprint compiler with a thin preset over the generic builder, removed unknown-owner fallback, classified helpers as exact supporting relationships, and deduplicated repeated per-surface semantics and oracles into stable model-owner obligations.
• Added independently reported qualification layers and an honest deepest proven layer plus the first incomplete layer. The static blueprint completion, user execution choice, implementation admission, and receipt-backed execution remain separate.
• Added a normalized shared-object projection and exact affected-neighborhood loader so ordinary work reuses owners, contracts, semantics, oracles, tests, resources, and intent without loading the whole blueprint or changing its logical identity when shard layout changes.
• Made affected loading validate only referenced objects and added per-model topology and model-test indexes, so unrelated shared objects are no longer a hidden whole-blueprint input to lightweight work.
• Added exact interrupted-owner settlement and separate standalone/child/parent evidence authority so a dead validation process can be closed without touching unrelated leases or treating progress as a terminal result.
• Updated the affected FlowGuard skill prompts and protocols so explicit whole-software work can reach blueprint depth while ordinary maintenance stays affected-only and lightweight.
• Used the current self-blueprint to perform a finite read-only, behavior-preserving architecture contraction audit. The current denominator is complete; all 11 size or repeated-shape signals remain risky_keep, with zero unproved automatic deletions or merges.
• Contracted the release-facing self-blueprint and architecture-reduction validators into one ten-owner self_maintenance_review producer, reused one exact in-memory blueprint, and replaced per-candidate full-surface caller scans with one deterministic reverse call index. Both validation obligations remain explicit; no cache or fallback was added.
• Made compact self-maintenance output a direct bounded projection instead of expanding a multi-gigabyte full graph and immediately discarding it.
• Reused the exact canonical fingerprint of immutable large behavior evidence across self-blueprint qualification and architecture reduction.
• Streamed canonical fingerprint and byte-count calculation for large normalized blueprint payloads, removing simultaneous multi-copy JSON amplification before physical projection construction.

v0.68.5 - 2026-08-02

• Made task facts independently observable by source plane and kept unknown, omitted, contradictory, unmapped, and scoped facts visible in the coverage denominator.
• Added one canonical owner-resolution identity across task coverage, model maturation, receipt verification, and readiness; caller-authored passed flags no longer substitute for command, result, time, subject, and content evidence.
• Added a read-only model-understanding-status API and CLI that reports model understanding, user execution choice, and FlowGuard implementation admission as three independent answers without running or writing anything.
• Replaced raw model-count activation with semantic topology and classified the complete 64-model observed universe into 7 parent domains with 189 explicit parent/consumer relations before whole-system closure.
• Contracted duplicate route, owner-result, Closure scoring, and count-based paths while preserving current public behavior, and updated the affected FlowGuard skills, contracts, tests, documentation, and release governance.
• Added an independent implementation and reconstruction-resource inventory, bidirectional model/code closure, source-independent semantics and oracles, and deterministic content-addressed blueprint projection. Static completion remains separate from optional empirical reconstruction, ordinary work stays affected-only, and no blueprint command launches reconstruction automatically.
• Added model-driven pre-release architecture reduction guidance that permits removal or collapse only with current equivalence evidence and keeps public delegation facades only when their exact owner and primary-path delegation evidence is current.
• Preserved content-addressed validation evidence under deep Windows worktrees by using extended-path reads and atomic writes.
• Removed user-specific absolute paths from the public model fixtures and archived coordination baseline before publication.

v0.68.4 - 2026-08-02

• Added an independently derived TaskCoverageDemand so task facts determine the minimum modeling owners and depth while callers may only add coverage.
• Added canonical maturation receipt publication and verification, then bound implementation admission, risk confidence, closure, and code structure to the same exact verified model identity.
• Materialized the upgraded understanding path in FlowGuard's own executable models, behavior commitments, affected topology, and model/test alignment.
• Simplified the ordinary AI route into a short facts-to-demand workflow while keeping deeper ModelMesh, Model-Test Alignment, TestMesh, and release rules reachable through explicit conditional protocols.

v0.68.3 - 2026-08-01

• Added task-local model-maturation intake that combines independently owned coverage and accepts resolved model angles only with exact current proof.
• Separated model sufficiency from implementation authorization: read-only work remains no-code, normal implementation requires closed current maturation, and explicitly bounded work can use a scoped authorization.
• Bound risk, closure, and code-structure readiness to the same maturation and admission identities so a complete-looking diagram cannot certify code by itself.
• Updated FlowGuard entry guidance and specialist protocols to report model depth and implementation admission separately while preserving lightweight use and target-software roles.

v0.68.2 - 2026-08-01

• Narrowed the ordinary FlowGuard entry to route selection and a minimum valuable model, while loading deeper modeling guidance only when the chosen route or an observed gap requires it.
• Replaced order-based route fallback with explicit positive, forbidden, and conflict conditions, and completed the public route table without changing the existing route ownership boundaries.
• Made prompt-budget checks derive the real local reference load graph and require usable headroom instead of treating an incomplete declared bundle as proof that the entry is lightweight.
• Tightened existing-model ownership projection to exact logical identity, normalized path identity, or matching fingerprints rather than basename-like guesses.
• Added a pointer-free model-revision-build handoff that accepts only an exact-current terminal-pass full model-regression parent receipt and emits typed content-addressed candidate/revision artifacts for the existing atomic activation command.

v0.68.1 - 2026-07-31

• Closed the clean-clone project-audit gap by keeping the adopted package version and generated project rules aligned with the released package.
• Kept every FlowGuard skill inside its frozen prompt budget by compressing the shared generated rules while preserving the protected behavioral guidance.
• Restored strict-validation parity with the CI-pinned OpenSpec 1.3.1 parser without weakening the current requirement semantics.

v0.68.0 - 2026-07-31

• Replaced the permissive model-maturation reader with one strict current-only task, coverage, probe, candidate, receipt, gap-lineage, and predecessor contract; former payload shapes now fail visibly.
• Made caller-authored resolution and self-reported understanding powerless: only exact current target-native producer receipts can resolve a gap.
• Added multi-iteration progress, stall, oscillation, limit, external-boundary, and non-vacuous closure checks plus executable positive and negative tests.
• Bound every maintained FlowGuard skill to an explicit target-owned model_deepening_check_id for SkillGuard supervision.

v0.67.0 - 2026-07-31

• Added task-local iterative model maturation with prediction, falsifier, gap-transition, progress, and explicit terminal records.
• Added a native CLI review route and regression coverage; matching local perturbations no longer close a task while addressable model gaps remain.

v0.66.0 - 2026-07-30

• Added a read-only model-miss diagnostic projection that derives deterministic deletion-minimal conflicts from exact observation, model, code/test, and failure-boundary atoms while preserving the existing Model-Miss Review owner decision.
• Made bounded incompleteness explicit when oracle budgets expire and attached per-atom necessity witnesses, stable ordering, and the diagnostic algorithm version instead of overclaiming a global or unique minimum.
• Added positive non-vacuity witnesses and repair-candidate checks that reject obligation deletion, unpreserved positive behavior, and missing intended rejection reasons before routing an accepted candidate to Model-Test Alignment.
• Extended the public API, model-miss skill guidance, executable native runner, focused tests, model-purpose closure, and observed model-system authority for the new evidence surface.
• Kept the release source-only: the immutable annotated tag is the release authority and the GitHub Release has zero uploaded assets.

v0.65.1 - 2026-07-28

• Restored strict-validation parity between the CI-pinned OpenSpec 1.3.1 parser and the current 1.6.0 parser for four wrapped requirement openings across three canonical specs without changing their requirement semantics.
• Preserved immutable v0.65.0 history and published this correction as a source-only patch release after both parsers passed all 99 current specs.

v0.65.0 - 2026-07-28

• Replaced broad, self-reported freshness with a frozen validation-owner DAG, resource-wide single-flight leases, supervised child processes, exact child receipts, and one independently verified full parent receipt.
• Made model authority inventory-complete and revision transactions independently derived, closure-complete, live-rechecked, CAS-protected, and rollback-replayable.
• Reconciled all 77 historical OpenSpec changes against current product direction, restored still-current requirements, retired obsolete topology and compatibility authority, and added byte-exact semantic sync checks.
• Added explicit complete/scoped UI evidence claims, package-authority shadow verification, behavior-bearing callable fingerprints, and StructureMesh child ownership for process supervision and the package API registry.
• Folded in the unreleased 0.64.2 validator-toolchain parity repair so current requirements parse under both the CI-pinned and current OpenSpec parsers.
• Kept the release source-only: the immutable annotated tag is the release authority and the GitHub Release has zero uploaded assets.

v0.64.2 - 2026-07-28

• Made the affected OpenSpec requirement statements valid under both the CI- pinned OpenSpec 1.3.1 parser and the current local 1.6.0 parser by placing the normative SHALL clause in each requirement's opening line.
• Kept the v0.64.1 runtime and release-authority behavior unchanged; this patch closes only the validator-toolchain parity gap exposed by GitHub CI.
• Preserved immutable v0.64.0 and v0.64.1 history; v0.65.0 incorporates this unreleased patch intent without moving either immutable tag.

v0.64.1 - 2026-07-28

• Closed a clean-clone authority gap by requiring the selected observed snapshot and every resolved model input to be Git-tracked before local release confidence.
• Added the 49 model and runner inputs that v0.64.0 referenced from the ignored local .flowguard/ tree but did not include in its committed source tree.
• Added observed missing-model and same-class missing-runner regressions and made published annotated-tag lookup safe across Windows command wrappers.
• Preserved immutable v0.64.0 history and published the correction as a source-only patch release after clean-clone project audit; CI later exposed the OpenSpec 1.3.1 parsing gap corrected in v0.64.2.

v0.64.0 - 2026-07-28

• Replaced unconditional full validation with a deterministic native owner plan whose rows are execute, reuse_current, or blocked, backed by immutable independently verified child receipts and single-flight leases.
• Preserved successful child evidence across parent failures so later runs execute only missing, failed, or stale owners and cheaply recompose the exact parent receipt.
• Split model-local identity from the global model-system snapshot revision and Git release provenance, preventing an unrelated model change or commit from invalidating unchanged sibling models.
• Added the v3 model regression impact map with explicit local/shared/ snapshot-only ownership and zero-producer fail-closed behavior for unknown, conflicting, malformed, or tampered inputs and receipts.
• Replaced mtime release freshness with separate content-addressed ValidationInputManifest and ReleaseTreeManifest identities, exact parent receipt verification, and immutable local-tag/published release checks.
• Direct-migrated the maintained FlowGuard skills, OpenSpec lifecycle, model authority, and source-only fifteen-skill consumer projection to the current contracts without compatibility readers or fallback execution paths.

v0.63.0 - 2026-07-27

• Made Existing Model Preflight select the same-plane owner closure before materializing model bodies, with distinct light/full projections and an explicit broad-authority mode.
• Added combined first-read prompt-bundle budgets and reduced the always-loaded project guidance while preserving each consumer skill's independent hard gates and route ownership.
• Added artifact-backed compact terminal envelopes for model and skill validation so successful child payloads and stream tails are not repeated, while failed, blocked, skipped, and not-run children remain visible.
• Required explicit behavior-source admission for WorkContext coverage, made Behavior Commitment Ledger discovery breadth follow its change mode, and prevented ordinary existing-ledger work from silently becoming bootstrap.
• Strengthened test-result reuse with independent producer terminal receipts, execution-owner matching, and exact producer/current fingerprints for command, source, artifact, dependencies, environment, result, and scope.
• Updated the existing FlowGuard models and maintained skill routes in place, retained source-only immutable release authority, and synchronized the clean fifteen-skill consumer projection.

v0.62.0 - 2026-07-26

• Replaced the OpenSpec-specific SpecContext surface with a provider-neutral, content-addressed, project-bounded read-only WorkContext registry and peer adapters for OpenSpec and explicitly declared files such as Spec Kit, Superpowers, custom skills, and ordinary planning documents.
• Added an independently derived exact expected-coverage inventory that joins WorkContext, observed UI, and field discovery before the Behavior Commitment Ledger assigns exactly one modeled, delegated, or scoped disposition.
• Strengthened UI and field completeness with immutable inventory revisions, discovery evidence, content fingerprints, exact ownership, field default / absence / serialization / privacy semantics, and conflicting-disposition rejection.
• Made TestMesh distinguish executed, delegated, and scoped coverage without treating provider status as test evidence, and made model-authority audits compare the selected snapshot with the current live manifest and exact owner/source inventories.
• Extended whole-system revision sets to account exact changed source-surface ids and every owner-artifact coverage dimension before atomic activation.
• Upgraded the existing FlowGuard skill routes and SkillGuard contracts in place, removed provider execution/session/cache/receipt bridges and retired SpecContext aliases, and retained source-only immutable release authority.

v0.61.0 - 2026-07-23

• Added one durable authoritative observed-model system snapshot with explicit separation between observed implementation, normative target, and isolated counterfactual experiment lanes.
• Added content-addressed model instances, typed cross-model relations, six coverage dimensions, exact subject revisions, and authority-first existing model preflight so discovery evidence can no longer masquerade as current authority.
• Added whole-revision-set activation and operational rollback with shared compare-and-swap locking, pointer-last persistence, affected-closure fingerprints, prediction/replay evidence, and required effect restoration or forward repair for irreversible changes.
• Integrated model authority into project audit, regression receipts, the command line, public API, behavior commitment ownership, and the existing FlowGuard skill routes without creating a parallel workflow.
• Split the former oversized authority implementation into schema, revision, persistence, and inventory owners; centralized the fixed skill-suite inventory; and removed the absent lightweight model-miss-review ghost route.
• Retained source-only release authority with an immutable tag and zero GitHub Release assets.

v0.60.0 - 2026-07-22

• Added strict flowguard.portable_system.v1 system definitions, verification requests, and exact declared-graph slices with separate canonical identities.
• Added bounded joint-system compilation and a stage-aware composition report that distinguishes pass, fail, blocked, invalid, and not_run, maps counterexamples back to components, and invokes the canonical system checker at most once.
• Added portable-system-check, Python API exports, and three benchmark families covering retry identity, permission revocation, and deletion propagation with bad, repaired, missing-semantics, and truncated variants.
• Extended the existing portable-composition behavior commitment, executable model evidence, regression manifest, tests, and affected FlowGuard skill guidance without adding a second runtime owner or a new satellite skill.
• Retained source-only release authority with an immutable tag and zero GitHub Release assets.

v0.59.0 - 2026-07-22

• Added one manifest-backed python -m flowguard simulator surface for model listing and explicit selected/all-model execution while preserving every model's native runner as its sole domain-result owner.
• Replaced full stdout/parsed-payload replication in model and unified release validation with deterministic compressed content objects, bounded diagnostics, hashes, and artifact references.
• Added terminal run manifests, atomic scoped current heads, read-only evidence audit/GC planning, exact quarantine/restore, and separately authorized purge; ordinary validation never silently cleans persistent evidence.
• Registered two previously unaccounted checkout-local models as optional_local, refreshed one stale model-purpose fingerprint, and extended the existing structure-surface model with simulator-takeover, duplicate- payload, and automatic-purge known-bad cases.
• Clarified that executable model source is small and distinct from generated evidence, build products, installed consumer skills, and extra local worktrees.
• Retained source-only release authority with an immutable tag and zero GitHub Release assets.

v0.58.5 - 2026-07-19

• Refreshed every FlowGuard skill's private author-side SkillGuard contract with the current affected-only validation ownership and provenance model.
• Kept graduated consumer skills clean: author contracts and receipts remain excluded from the installed projection, with no fallback or compatibility runtime.
• Retained source-only release authority with an immutable tag and zero GitHub Release assets.

v0.58.4 - 2026-07-19

• Corrected the v0.58.3 artifact-ownership model miss: a numeric schema_version, producer label, FlowGuard-looking prefix, scan path, or partial ledger shape no longer grants FlowGuard write authority.
• Added an exact current-only registry for real FlowGuard report and trace envelopes. Current shapes remain byte-identical; malformed, old, future, and unsupported versions block without writing because no evidence-bound migrator exists.
• Retained one repository-history-backed behavior-ledger writer, bounded to the exact complete bare producer shape from commit 56083c1e. It materializes the full current canonical envelope at the upgrade boundary; target-only extra fields and old/future BCL envelopes never enter it.
• Required explicit upgrade-AI plane/actor dispositions in legacy row metadata; actor, owner, label, trigger, and commitment-kind text are never used to guess new current semantics.
• Migrated historical primary_path_ids explicitly at that boundary: zero or one value becomes the current singular field, while multiple values block for an evidence-bound disposition. No generic version-field writer, fallback, dual reader, or compatibility parser was added.
• Removed the normal-runtime plural path reader and its legacy_plural_migrated / primary_path_migration_ambiguous state. The BCL loader now requires the exact current envelope field set, current schema and format versions, and a canonical current payload.
• Updated the bundled public BCL template to emit that complete current canonical payload directly; template loading never fills omitted fields or consults a compatibility reader.
• Added a real non-editable-package regression that migrates a historical FlowGuard behavior ledger while proving a Khaos-style target fixture remains byte- and SHA-256-identical in the same project-upgrade run.
• Extended both public upgrade models with target-owned JSON, full legacy-lookalike extra-field, and unsupported-envelope known-bad cases.
• Retained source-only release authority with an immutable tag and zero GitHub Release assets.

v0.58.3 - 2026-07-19

• Corrected the v0.58.2 deployment-topology miss: ordinary project audit and upgrade no longer look for the author-only suite map through an editable package layout.
• Added one deterministic package-owned clean-consumer authority containing the exact 15 member ids and complete projected file fingerprints, including generated consumer-release manifests.
• Made ordinary project writes require exact package-authority parity with the global $CODEX_HOME/skills/ projection and its ownership manifest. Missing, modified, reserved-extra, or author-control-polluted consumers block visibly.
• Added a real non-editable package regression that installs into an isolated target directory, upgrades a genuinely empty ordinary project, verifies the expected writes, and passes the subsequent project audit.
• Kept author suite maps, SkillGuard contracts, receipts, and check execution outside the installed consumer and ordinary project runtime. No fallback, alternate reader, compatibility path, or project-local shadow suite was added.
• Retained source-only release authority with an immutable tag and zero GitHub Release assets.

v0.58.2 - 2026-07-18

• Repaired an ordinary-project authority regression: generated minimum revalidation now invokes only the installed package audit and no longer requires the author-repository verify_skill_suite_markers.py script.
• Made the one clean consumer projection under $CODEX_HOME/skills/ the sole FlowGuard skill authority for ordinary projects; adopted projects do not receive a second .agents/skills tree or suite map.
• Added executable temporary-project regressions proving project adoption, audit, and minimum revalidation work without local scripts, .agents/skills, or .skillguard directories.
• Preserved the author-maintenance boundary: the canonical suite map, full fifteen-member SkillGuard validation, and generated author contracts remain owned by the FlowGuard source repository.
• Kept one direct current path only; no alias, fallback, compatibility reader, or shadow-suite authority was added.
• Retained source-only release authority with an immutable tag and zero GitHub Release assets.

v0.58.1 - 2026-07-18

• Corrected a v0.58.0 model miss in project audit and upgrade: suite reconciliation now uses the canonical map from the current FlowGuard package source and the clean consumer projection from the current installed Codex skills root.
• Added a regression proving ordinary projects need no .skillguard directory or local FlowGuard skill suite, while missing or mismatched installed consumer skills still block before mutation.
• Retained one direct current path only; no target-local compatibility reader, alias, or fallback was added.
• Removed the Python 3.10 hand-written TOML parser fallback and made Python 3.11+ with the standard-library tomllib parser the sole current runtime path.
• Removed the unused run_flowguard_self_governance.py forwarding alias; the sole current command is check_flowguard_self_governance.py.
• Removed stale regression-manifest registrations for the already retired SpecContext and task-local replay models and refreshed the changed ModelMissReview purpose identity.
• Made CI materialize the current clean consumer skill projection before the project-adoption audit, so a clean runner follows the same declared install path instead of depending on undeclared machine state.

v0.58.0 - 2026-07-18

• Renamed the canonical kernel skill directly from model-first-function-flow to flowguard; no alias, forwarding stub, compatibility reader, or fallback path remains.
• Reduced the public Codex surface from seventeen to fifteen skills by moving plan detailing and agent-workflow rehearsal into two explicit internal routes owned by flowguard-development-process-flow.
• Consolidated all fifteen author sources into the single unit:flowguard-suite SkillGuard maintenance unit while preserving each public satellite's native route and check authority.
• Made source tags the sole release authority: v0.58.0 releases contain no wheel, source-distribution, or downloadable package assets.
• Added explicit author-source and consumer-distribution parity roles so consumer projections are compared without author-only SkillGuard material.

v0.57.0 - 2026-07-16

Portable compositional-verification kernel release.

• Added a strict, versioned, JSON-serializable finite model IR with canonical identities, explicit invariants and temporal obligations, and fail-closed rejection of unknown or dangling schema content.
• Added a code-independent reference checker for nondeterministic traces, safety, terminal progress, universal and bounded eventuality, weak fairness, visible truncation, and smallest counterexample preservation.
• Added explicit child-to-parent refinement and assume/guarantee composition checks, including state/transition mappings, stutter declarations, assumption weakening, guarantee strengthening, provider closure, and conflict detection.
• Exposed one narrow Python API and CLI cohort, registered the kernel in the model-regression manifest, and updated ModelMesh, topology-hazard, and model-first skill contracts without adding a duplicate public route.
• Revalidated the project artifacts with official OpenSpec 1.6.0. This local release deliberately excludes GitHub push, remote tag, and GitHub Release evidence.

v0.56.0 - 2026-07-15

Guard-purpose closure and generic declared-check supervision release.

• Added executable model-purpose contracts so every FlowGuard model declares the concrete bad behavior it is intended to block for the current modeling task, the accepted boundary, observable evidence, and the safe claim that a passing model licenses.
• Propagated purpose evidence through model review, model/test alignment, TestMesh, development-process closure, and the complete seventeen-skill agent suite without turning SkillGuard into a domain-semantics owner.
• Replaced former SkillGuard authorities with the sole current generic declared-check contract shape: exact target-owned checks, one execution owner per check, current immutable receipts, fixed enforced closure, and no selectable quality or integration mode.
• Hardened multi-skill validation ownership, installation projection, OpenSpec receipt reuse, stale-evidence rejection, interrupted-process cleanup, and source/toolchain identity binding.
• Updated models, contracts, templates, documentation, examples, fixtures, OpenSpec work packages, and regression coverage for the direct-current release boundary.

v0.55.0 - 2026-07-12

FlowGuard product-language, behavior-plane, and validated-path reuse hardening.

• Partitioned behavior commitment lookup by execution plane while preserving typed actor/relationship bindings and canonical ledger I/O, so plane-aware discovery composes with existing commitment ownership instead of replacing it.
• Added stable exact business-intent identity, one active commitment per exact external promise, one singular selected primary path, deterministic one-item legacy plural migration, complete affected-surface/candidate inventories, and material current runtime/proof requirements.
• Extended the existing UI Flow Structure owner with product-scope typography, component, navigation, interaction, feedback, recovery, and transition consistency; repeated business surfaces bind the same commitment/path, while pure UI interactions stay local.
• Kept UI content admission exactly user_visible, user_on_demand, or internal; internal intent, commitment, path, audit, evidence, and diagnostic ids remain outside ordinary UI.
• Hardened RuntimePathEvidence, ObligationFamily, Model-Test Alignment, Architecture Reduction, ContractExhaustionMesh, TestMesh, and DevelopmentProcessFlow against omitted inventories, opaque handoff ids, sibling-obligation evidence, parallel facade success, stale proofs, progress-only background runs, cross-domain sync overclaims, and peer-write rollback.
• Updated OpenSpec requirements, existing FlowGuard models, tests, templates, public API guidance, skill prompts/contracts, local installation parity, and non-destructive shadow/formal/Git synchronization evidence without adding a product-language route, path-reuse route, intent ledger, or evidence engine. Added the read-only behavior-commitment-query command under the existing Behavior Commitment Ledger / Existing Model Preflight owners; it is not a new FlowGuard route and does not execute the retrieved commitment.
• Migrated all seventeen bundled skills to skillguard.contract_source.v2 with existing-owner model projections, generated V2 contracts/manifests, and contract_mapped depth profiles. Static contract-depth certification remains distinct from actual execution-depth receipts and adds no parallel executor.

v0.54.1 - 2026-07-10

FlowGuard published-release verification compatibility patch.

• Resolve external commands through the operating system executable search so Windows PATHEXT shims such as git.CMD work in published verification.
• Convert missing external commands into structured failed checks instead of terminating the verifier with an unhandled process-launch exception.
• Correct annotated-tag verification by comparing the local release commit with the remote tag's peeled commit while retaining lightweight-tag support.

v0.54.0 - 2026-07-10

FlowGuard skill-suite governance and verified distribution release.

• Rebuilt the canonical seventeen-skill agent suite around concise, route-specific prompts, typed ownership, deterministic contract sources, generated native-integrated SkillGuard records, and 17/17 deep checks.
• Added immutable evidence receipts whose freshness, covered obligations, producer/checker context, child consumption, and proof-artifact bindings are independently recomputed instead of accepted from caller-supplied flags.
• Hardened FlowGuard self-governance so broad success requires current exact child passes, a complete parent receipt, and visible blocked/scoped/stale or skipped states; synthetic in-memory green reports no longer satisfy closure.
• Added a typed route topology with unique owners, bounded cycles, route-parity checks, and explicit delegated/internal/public handoffs.
• Productized local model validation with a checked-in inventory, fast, focused, and full tiers, filters, shards, bounded parallelism, per-runner timeouts, mutation detection, progress events, isolated artifacts, and terminal receipts.
• Added safe, idempotent skill install/check/uninstall/parity commands with ownership manifests, explicit legacy adoption, user-change conflict protection, temporary-home tests, and complete source/shadow/installed tree comparison.
• Repaired project adoption integrity, stable rule identities, dry-run guarantees, concise validation output, bilingual documentation, and OpenSpec verification contracts for the complete upgrade program.

v0.53.1 - 2026-07-09

FlowGuard self-maintenance gate hardening.

• Added strict Behavior Commitment Ledger self-maintenance checks so FlowGuard workflow changes must keep behavior registration, DCAR coverage, TestMesh routing, and model-miss backfeed connected before broad confidence.
• Extended behavior commitment records with source freshness, replacement synchronization, TestMesh alignment, model-miss status, and change-mode handling for new, changed, replacement, discovery, and model-miss work.
• Updated FlowGuard agent skills, templates, public APIs, CLI rendering, and tests so missing ledger coverage blocks immediately instead of leaving an alternate success path.

v0.53.0 - 2026-07-08

Behavior Commitment Ledger, Primary Path Authority, and no-fallback coverage.

• Added behavior_commitment as the upstream Behavior Commitment Ledger route so broad work inventories every external, verifiable behavior commitment before assigning one primary owner model and downstream evidence.
• Connected path-sensitive behavior commitments to primary_path_authority, so the ledger owns full behavior registration while PPA owns single primary path authority and rejects alternate success after primary failure.
• Added primary_path_authority as a first-class FlowGuard route so each path-sensitive business intent declares one primary runtime authority and rejects automatic alternate success after primary failure.
• Wired no-fallback evidence through RuntimePath observations, RiskEvidenceLedger gates, public API/CLI/template surfaces, AGENTS guidance, and installed FlowGuard skill shells.
• Added model-scoped Cartesian coverage for the primary-path boundary with actionable oracle binding, TestMesh shard ownership, and downstream risk evidence consumption.
• Published the previously untagged OpenSpec readiness verification commit and included validated OpenSpec records for ContractExhaustionMesh, hierarchical Cartesian coverage, and helper control-plane consolidation.

v0.52.6 - 2026-07-07

Guard-family simulation readiness patch.

• Restored route-specific diagram intent guidance in the model-first kernel so downstream Guard skills are not flattened into generic flowcharts.
• Hardened FlowGuard SkillGuard work contracts so duplicate SkillGuard-owned execution paths are invalid while native FlowGuard checks remain authoritative.
• Added OpenSpec release tracking and source-side regression coverage for the prompt contract.

v0.52.5 - 2026-07-06

Model-code-test closure hardening.

• Added first-class optional source-audit gating to Model-Test Alignment for real-code claims, including missing, failed, and unrelated-audit blockers.
• Expanded ModelCodeTestBindingRow into a behavior closure summary with owner code, path/symbol, tests, boundary/runtime/payload ids, source-audit decision, and open gap codes.
• Added target-aware counterexample and known-bad replay closure through external owner-code-bound TestEvidence.
• Added structured runtime writer-inventory evidence for Runtime Gateway Adoption and updated public docs, templates, skills, and tests.

v0.52.4 - 2026-07-03

FlowGuard skill-suite distribution.

• Reframed public onboarding around .agents/skills/ as the primary AI-agent skill-suite surface, with model-first-function-flow as the default entrypoint and sibling FlowGuard skills kept visible.
• Clarified that the Python flowguard module and CLI are executable check-engine support, not the AI-agent skill installation surface.
• Added skill-suite marker verification and updated project adoption templates, docs, and tests so repository, installed Codex skills, and generated AGENTS guidance stay aligned.
• Archived the reframe-flowguard-as-skill-suite OpenSpec change into the main specs and refreshed release-scope evidence.

v0.52.3 - 2026-06-27

• Added SkillGuard runtime-contract governance for the installed FlowGuard Codex skill materials.
• Synchronized installed skill copies with accepted source material and local git evidence.
• Recorded release-scope validation so route selection, evidence gates, quality floors, and closure boundaries remain visible before completion claims.

v0.52.2 - 2026-06-25

FlowGuard current-situation visibility.

• Added lightweight current-situation guidance so non-trivial FlowGuard work tells users what is being checked, why it matters, current evidence or gaps, and the next step before or with Mermaid snapshots.
• Updated the shared agent snippet, model-first kernel, and direct FlowGuard satellite skills with concise status-note wording while preserving hot-path prompt budgets and existing route ownership.
• Archived the OpenSpec change into flowguard-model-visibility and user-facing-model-diagrams, and refreshed local model checks so diagram-only prompt rollouts without current-situation notes are rejected.

v0.52.1 - 2026-06-20

FlowGuard self-maintenance and release hygiene.

• Archived the completed ContractExhaustionMesh OpenSpec changes into the main specs so the active change list is clean before publication.
• Re-ran FlowGuard self-maintenance, DevelopmentProcessFlow, Model/Test/Code alignment, TestMesh, and StructureMesh checks and kept the current route structure unchanged because no model-backed split or field deletion was required.
• Prepared a source-only patch release that aligns public version metadata, project adoption records, changelog, README, Git tag, and GitHub Release.

v0.52.0 - 2026-06-20

Contract coverage universe and observed-problem backfeed.

• Added ContractCoverageUniverse, scoped exclusions, and broad-claim gates so ContractExhaustionMesh can prove the declared coverage universe is present instead of only proving the generated matrix is internally consistent.
• Added generic ContractFaultProfile and ObservedProblemBackfeed helpers so downstream systems can rehearse bad contract submissions and map real misses back to generated, same-class, and coverage-receipt evidence without a FlowPilot-specific fake-AI API.
• Updated route profiles, skills, docs, templates, OpenSpec change records, API exports, and regression tests for actionable oracle feedback and universe completeness.

v0.51.0 - 2026-06-16

Hierarchical Cartesian contract-exhaustion coverage.

• Added model-scoped axes, interaction groups, generated combination cases, coverage shards, and coverage receipts so ContractExhaustionMesh can produce finite bad-case matrices per model instead of relying on hand-written representative examples.
• Wired generated combination obligations into Model-Test Alignment, TestMesh, ModelMesh, RiskEvidenceLedger, bug-family records, and model-miss review so a green single-point matrix cannot be mistaken for a closed parent/child model chain.
• Updated FlowGuard route profiles, skills, templates, OpenSpec tasks, docs, topology, and regression tests so future AI agents see the Cartesian coverage checklist and required handoffs before claiming completion.

v0.50.0 - 2026-06-15

Helper control-plane consolidation.

• Added first-class route-role metadata to RouteProfile so public owner routes, delegated modes, internal feeders, data helpers, and cleanup dispositions are machine-checkable instead of prompt-only guidance.
• Split public route discovery from advanced/internal helper inventories so FLOWGUARD_ROUTE_API and ROUTE_STARTER_API expose owner routes while simulator, model-angle, similarity, maintenance-scan, closure, and state-closure helpers remain owner-consumed evidence.
• Updated FlowGuard skills, AGENTS guidance, API docs, topology docs, project adoption generation, self-maintenance models, and regression tests so DevelopmentProcessFlow, ExistingModelPreflight, ContractExhaustionMesh, ArchitectureReduction, and RiskEvidenceLedger own the single hot path without helper-first fallback routes.

v0.49.0 - 2026-06-15

Development-process simulator entry consolidation.

• Added development_process_simulator helpers so rough-plan discussion, multi-skill workflow setup, staged execution, install/sync, release/archive, publish, and final process claims enter flowguard-development-process-flow first with ordered plan_detailing, agent_workflow, and execution_freshness modes.
• Reframed PlanDetailing Compiler and AgentWorkflowRehearsal as explicit or simulator-delegated mode owners instead of competing generic first entries.
• Updated Codex skill prompts, OpenAI metadata, AGENTS guidance, README, API docs, current OpenSpec specs, and active plan-discussion change notes to remove stale direct/peer route wording.
• Added routing/model/API/docs regression coverage so old rough-plan and multi-skill hot-path wording is rejected.

v0.48.0 - 2026-06-15

Single formal FlowGuard model entry.

• Added structured KnownBadProof review so new or deepened models must prove representative broken cases are caught before broad FlowGuard claims.
• Hardened RiskIntent, MinimumModelContract, template reuse/no-match, and runner summary gates so missing protected errors, completion evidence, known-bad proof, or template closure blocks instead of silently passing.
• Removed direct Explorer from the agent-default API, migrated public templates and model-first skills to FlowGuardCheckPlan + run_model_first_checks(...), and added adoption audit warnings for direct-Explorer-only current models.

v0.47.3 - 2026-06-15

ModelMesh closure and retry/rejection liveness hardening.

• Made parent/child ModelMesh confidence require closure modeling whenever child outputs, reattachment contracts, or runtime path evidence are present.
• Added repeated-input retry/rejection closure fields, repair-feedback and blocker/progress checks, ModelMesh closure transition projection into Model-Test Alignment/TestMesh, and regression tests for stuck-loop hazards.
• Updated skills, templates, docs, OpenSpec coverage, field inventory, public API exports, and template/CLI known-bad proof wiring so current evidence, tests, and installed skills agree on the strengthened contract.

v0.47.2 - 2026-06-13

UI functional capability coverage.

• Added UI functional capability inventories, capability/output contracts, capability bindings, and coverage review so required user-visible UI functions cannot disappear before broad UI completion claims.
• Wired capability coverage into implementation validation, human-operability, PlanDetailing, Model-Miss Review, DevelopmentProcessFlow, RiskEvidenceLedger, AgentWorkflowRehearsal, and ClosureContract without creating a parallel UI route.
• Updated UI skills, public templates, docs, API/field inventory, OpenSpec artifacts, and regression tests so greenfield, source-based, and mixed UI work use the same generic capability layer.

v0.47.1 - 2026-06-13

Plan discussion handoff hardening.

• Promoted PlanDetailing as the direct FlowGuard route for non-trivial plan, acceptance, rough-outline, and AI-generated workflow discussions before execution routes consume the plan.
• Added AgentWorkflowRehearsal completion-ledger report fields for planned, completed, blocked, skipped, required-recheck, handoff, and final-claim boundaries.
• Updated skill routing docs, public templates, OpenSpec coverage, API/field docs, and regression tests so long prose plans no longer count as completion or release evidence.

v0.47.0 - 2026-06-13

Generic UI source-baseline validation.

• Added generic greenfield, source_based, and mixed UI work-mode modeling plus source baseline, target mapping, approved difference, observed-source alignment, and source interaction review helpers.
• Replaced source-specific UI migration wording, APIs, process/risk gates, closure evidence, agent-workflow roles, public templates, docs, and active OpenSpec changes with generic source-baseline terminology.
• Added regression coverage so active skills, templates, docs, and OpenSpec guidance do not hard-code a specific source UI technology while still allowing historical archive records.

v0.46.0 - 2026-06-12

Mandatory template harvest closure.

• Added TemplateHarvestReview, accepted harvest dispositions, not-harvestable reasons, and review_template_harvest_closure(...).
• Extended check plans, audit, runner summaries, CLI, public API exports, and starter templates so new or materially deepened models must close template harvest as written, merged, duplicate-linked, or accepted not-harvestable.
• Updated FlowGuard skills, AGENTS snippet, modeling protocol, README, API docs, OpenSpec coverage, and self-model evidence so template absorption is a hard completion gate rather than optional "when useful" guidance.

v0.45.0 - 2026-06-12

UI human-operability gate.

• Added user task coverage, region semantics, affordance contracts, action grammar, dialog/window return contracts, keyboard/focus contracts, human walkthroughs, and review_ui_human_operability(...).
• Extended UI Flow Structure, Model Miss Review, PlanDetailing, DevelopmentProcessFlow, RiskEvidenceLedger, ClosureContract, and AgentWorkflowRehearsal so broad UI done/release confidence requires human-operability evidence, not only labels, routes, or click wiring.
• Updated public templates, API docs, UI docs, README, OpenSpec artifacts, and skill guidance so supported UI tasks must connect to function model features, UI journeys, primary controls, functional chains, feedback, cancel/error, keyboard/focus, native dialogs, and walkthrough evidence.

v0.44.0 - 2026-06-12

Minimum valuable model entry and risk-template reuse.

• Added packaged public risk templates plus a portable per-machine local risk template library with search, review, merge, and harvest helpers.
• Raised the default AI model-first entry from a thin starter to a minimum valuable model that names protected errors, state/side effects, completion evidence, and known-bad cases.
• Updated RiskIntent, CheckPlan, audit, runner, model similarity, CLI, templates, skills, docs, and OpenSpec coverage so template reuse and local candidate absorption become part of normal model creation.

v0.43.3 - 2026-06-12

UI real-surface validation hardening.

• Added observed real UI inventory, enabled-control functional chain, MATLAB baseline callback semantics, and UI model-miss review helpers.
• Added PlanDetailing, DevelopmentProcessFlow, RiskEvidenceLedger, ClosureContract, and AgentWorkflowRehearsal gates so UI completion requires typed current evidence and final done-claim review.
• Updated OpenSpec specs, FlowGuard skills, docs, templates, and API surface so labels, routes, planned evidence, and API existence cannot stand in for real UI click-through behavior.

v0.43.2 - 2026-06-11

Artifact payload real-flow evidence tightening.

• Tightened Model-Test Alignment artifact payload validation so current external passing payload evidence must include an execution proof reference or proof artifact for the real payload surface.
• Updated FlowGuard skills, protocols, docs, and templates to describe synthetic payload cases as inputs for real import/export/save/load/generate or AI work-package surfaces, not standalone payload-only paths.
• Added regression coverage for declaration-only payload evidence and refreshed local sync guidance for repository, Gate, and installed Codex skill surfaces.

v0.43.1 - 2026-06-11

Agent workflow rehearsal completion ledger upgrade.

• Added planned_steps, completed_steps, blocked_steps, skipped_steps, required_rechecks, handoff_points, and final_claim_boundary to AgentWorkflowRehearsalReport.
• Updated the AgentWorkflowRehearsal skill and protocol so multi-skill plans expose what is planned, what is blocked, what was skipped, which evidence is handed off, and why rehearsal is not execution proof.
• Added focused regression coverage for the new report fields.

v0.43.0 - 2026-06-09

Writing quality-gate process upgrade.

• Added DevelopmentProcessFlow guidance for thesis and source-backed writing quality-gate evidence, including literature/technology progression, method-depth, figure/table treatment, citation/footnote support, source/trace handoff, AI-style integration, final prose, and revision-report freshness.
• Updated PlanDetailing and AgentWorkflowRehearsal skills so large writing plans and multi-skill workflows expose these quality gates before execution or done/release confidence.
• Updated public release metadata for the new process-surface guidance.

v0.42.0 - 2026-06-08

Validation evidence gate upgrade for UI and artifact payload claims.

• Added Model-Test Alignment artifact payload helpers for file import/export, generated artifact, save/load, and AI work-package validation cases.
• Integrated payload contracts/evidence into ModelTestAlignmentPlan and the public API surface so missing, stale, internal-only, manual prose-only, or mismatched payload evidence blocks broad alignment claims.
• Added Risk Evidence Ledger gates for UI implementation evidence and artifact payload validation before done/release/full-confidence claims.
• Strengthened UI, Model-Test Alignment, TestMesh, DevelopmentProcessFlow, PlanDetailing, AgentWorkflowRehearsal, kernel guidance, docs, and templates so runnable UI claims require reachable control click-through evidence and payload/work-package claims require synthetic payload packs.

v0.41.8 - 2026-06-07

AI-entry surface reduction for thinner default FlowGuard authoring.

• Added ROUTE_STARTER_API, ROUTE_ADVANCED_API, PLAN_INTAKE_STARTER_API, and PLAN_INTAKE_ADVANCED_API so agents can read compact route-specific entry points before full helper inventories.
• Made model-miss, model-test-alignment, and UI-flow template commands emit compact runnable defaults while adding explicit full-template commands for deep scaffolds.
• Added AI surface tier and route-owner metadata to the generated field lifecycle inventory so later field deletion can start from starter, advanced, and internal exposure evidence.
• Added a FlowGuard self-check model for AI-entry reduction and updated docs, skills, and tests to keep starter surfaces compact without hiding full escalation paths.

v0.41.7 - 2026-06-07

Breaking field-schema cleanup for thinner AI-authored route inputs.

• Replaced route-specific RiskEvidenceRow gate columns with a compact RiskEvidenceGate list and updated ledger checks, docs, templates, tests, and local FlowGuard models to reject old field shapes.
• Removed AutoSplit metric fields from ProcessEvidence; split review now stays in AutoSplit, ModelMesh, or TestMesh evidence instead of being copied into development-process rows.
• Removed duplicate PlanIntake mapping/source evidence shapes and adapter fixture flags from normal mapping rows.
• Merged duplicate UI blindspot helper classes into UIBlindspot and reused ProcessArtifact in PlanDetail instead of maintaining a duplicate PlanDetailArtifact class.

v0.41.6 - 2026-06-07

FieldLifecycleMesh route-evidence handoff binding.

• Added minimal gate:, test:, and replay: evidence-route checks for broad behavior-bearing field lifecycle claims while keeping bounded plans backward compatible.
• Updated the default replacement field-lifecycle model so full replacement claims require a route-bound field projection before disposal can close.
• Updated public templates, docs, tests, and installed skill guidance to make field models behave like route tracking instead of a field dictionary.

v0.41.5 - 2026-06-07

Deep CI portability repair after v0.41.4.

• Fixed the shadow workspace sync helper so ignored project directories are evaluated relative to the source set instead of treating Linux system /tmp as a project tmp directory.
• Made the local closure-contract model-runner test skip gracefully when a GitHub checkout does not include ignored local .flowguard model artifacts.
• Kept the v0.41.4 maintenance hardening release intact and superseded it with this clean patch release instead of moving a public tag.

v0.41.4 - 2026-06-07

Complete FlowGuard maintenance hardening.

• Added AGENT_DEFAULT_API as a compact first-read surface for agents while preserving the full route and helper API registry.
• Added tracked maintenance helpers for aggregate local .flowguard model regression and no-delete shadow workspace synchronization.
• Added a generated field lifecycle inventory so field reduction can start from ownership and lifecycle clues instead of text-search deletion.
• Split Model-Test Alignment Python source-audit execution into a dedicated module while preserving the original public imports.
• Split GitHub CI into fast push checks and scheduled/manual deep checks that include full unit and aggregate model regression.

v0.41.3 - 2026-06-07

GitHub CI portability fix.

• Fixed artifact upgrade scanning so repository-relative ignored folders such as tmp are skipped without accidentally skipping Linux system temp roots.
• Kept public template privacy checks focused on real private markers while avoiding GitHub Actions' generic runner home directory false positive.
• Added regression coverage for the temp-root scan behavior exposed by CI.

v0.41.2 - 2026-06-07

Post-release maintenance hygiene.

• Archived completed OpenSpec changes into the durable archive and kept the active change list focused on current work.
• Added a compact default_flowguard_self_maintenance_plan() helper so agents can start FlowGuard self-maintenance with route, API, AI-entry, and field defaults instead of manually filling the common plan fields.
• Added a minimal GitHub Actions CI gate for install, project audit, OpenSpec strict validation, the self-maintenance model, and focused unit tests.

v0.41.1 - 2026-06-07

FlowGuard self-maintenance optimization.

• Added a route-first FlowGuard self-maintenance mesh with compact route profiles, AI entry profiles, field layer profiles, and child closure reports so agents can maintain FlowGuard without starting from the full helper list.
• Expanded public route grouping and API-surface tests so installed satellite routes are discoverable through stable route entries instead of scattered flat fields.
• Added OpenSpec coverage, a self-maintenance model, skill/docs updates, focused regression evidence, and shadow-workspace sync checks for the new maintenance path.

v0.41.0 - 2026-06-07

Open-ended model-angle deliberation.

• Added ModelAngleDeliberation, ModelAngleReviewReport, and review_model_angle_deliberations() so agents must record what a current model sees, what it misses, what fails if ignored, and whether to reuse, extend, add a child model, create a new model, scope/defer, or ask for human review.
• Connected unresolved model-angle gaps to Existing Model Preflight, Maintenance Scan, Risk Evidence Ledger, and Closure Contract so broad confidence cannot hide a missing model viewpoint.
• Added model-angle-template, API/docs/skill guidance, OpenSpec coverage, focused tests, and FlowGuard self-model evidence for the new route.

v0.40.12 - 2026-06-07

Guard-family closure contract upgrade.

• Added a reusable Guard closure contract checker for child Guard skills so passing claims cannot hide hard findings, missing inputs, stale evidence, or skipped checks.
• Updated the model-first skill guidance so SourceGuard, TraceGuard, LogicGuard, PhysicsGuard, research, and thesis workflows can hand AI agents structured next actions instead of relying on prose-only reminders.
• Added OpenSpec and FlowGuard self-model evidence for the closure-report maintenance path.

v0.40.11 - 2026-06-06

Default replacement and field lifecycle hardening.

• Added FieldLifecycleMesh helpers, route-scoped API discovery, a public field-lifecycle-template, and a Codex skill route for complete leaf-level field accounting.
• Connected behavior-bearing field projections and old-field disposition to existing model preflight, code structure, architecture reduction, Model-Test Alignment, Model-Miss Review, DevelopmentProcessFlow, legacy path disposition, and Closure Contract.
• Updated project AGENTS guidance, docs, public templates, OpenSpec coverage, skill protocols, and FlowGuard self-model checks so replacement work cleans up old paths/fields by default unless compatibility is explicit and evidenced.

v0.40.10 - 2026-06-05

AI route handoff optimization.

• Added structured owner-route, action, proof-gap, required-input, and claim boundary fields to SummaryReport findings, maintenance obligations, maintenance scan actions, and DevelopmentProcessFlow revalidation guidance.
• Added maintenance_scan_plan_from_summary_report(...) and existing_model_preflight_from_project(...) so agents can continue through existing FlowGuard routes instead of inventing a parallel session runner.
• Refreshed templates, compact skill guidance, API docs, OpenSpec artifacts, tests, and FlowGuard self-model evidence for the SummaryReport -> MaintenanceScan -> specialist-route handoff.

v0.40.9 - 2026-06-04

Bug repair closure hardening.

• Broadened Model-Miss Review so non-trivial bug repairs use the existing model-miss route instead of falling back to implementation-only patches.
• Added root-cause backpropagation, owner code-contract binding, same-class evidence, legacy path disposition, DevelopmentProcessFlow freshness, and Risk Evidence Ledger closure requirements to the bug repair loop.
• Refreshed the public model-miss template, self-model, OpenSpec change, focused tests, and docs so model, code, and tests stay tied to the same repaired behavior before broad confidence.

v0.40.8 - 2026-06-03

Soft UI typography hierarchy guidance.

• Clarified that UI text hierarchy blueprints are semantic handoff artifacts, not instructions to create one visual font size per hierarchy level.
• Updated FlowGuard UI Flow Structure guidance, templates, docs, and focused tests so typography noise is treated as a design smell with intentional exceptions rather than a hard numeric rule.
• Synced first-pass design, implementation review, and design iteration skills with calm hierarchy, reusable text treatment, and clear-rationale guidance.

v0.40.5 - 2026-06-02

Maintenance obligation memory.

• Added route-owned MaintenanceObligation memory helpers so unresolved structure, evidence, model, and risk gaps can stay visible without creating a standalone technical-debt scanner.
• Extended summary reports, maintenance scans, model maturation, and Risk Evidence Ledger reviews so anchored open obligations reopen owner routes and broad confidence is blocked or scoped until obligations are resolved.
• Updated public templates, OpenSpec coverage, FlowGuard skills, docs, version records, and self-model checks for the maintenance-obligation memory flow.

v0.40.4 - 2026-06-01

Model topology hazard review.

• Added review_topology_hazards() with UsageIntent, topology digests, anchored hazard candidates, confidence decisions, runner integration, and route-scoped API discovery.
• Added automatic topology_hazard summary sections to run_model_first_checks(...) so model-shape future-use hazards stay visible before broad done, release, publish, archive, or full-confidence claims.
• Connected topology hazard gaps to maintenance scan, Risk Evidence Ledger, public templates, OpenSpec artifacts, docs, Codex skill guidance, and FlowGuard self-model checks.

v0.40.3 - 2026-06-01

Latest-schema-first upgrade policy.

• Added deterministic artifact upgrade scanning for old FlowGuard artifacts, model/test evidence, docs, scripts, and skill guidance.
• Extended project-upgrade so older adopted repositories upgrade known old artifacts by default, with --records-only for narrow record refreshes.
• Added the artifact-upgrade CLI, public report helpers, OpenSpec coverage, self-model checks, focused tests, and updated agent guidance.

v0.40.2 - 2026-06-01

Automatic state/input closure gate.

• Added review_state_closure() with typed closure dimensions, representative unknown cases, safe handling policies, reports, and route-scoped API discovery.
• Added automatic state_closure summary sections to run_model_first_checks(...) so inferred unknown/other cases scope confidence by default and unsafe handling blocks confidence.
• Routed state closure maintenance gaps to model_maturation_loop and updated docs, skill guidance, OpenSpec artifacts, tests, and the FlowGuard self-model.

v0.40.1 - 2026-06-01

FlowGuard field prompt reduction.

• Collapsed Model-Test Alignment, DevelopmentProcessFlow, TestMesh, ModelMesh, and adoption-log prompt fields into grouped field families.
• Preserved required evidence, freshness, boundary, skipped/not-run, release, and scoped-gap fields while moving rare details behind expand-when-applicable guidance.
• Added skill documentation tests and a FlowGuard self-model to prevent long blank field lists or evidence-dropping reductions from returning.

v0.40.0 - 2026-06-01

FlowGuard maintenance scan router.

• Added review_maintenance_scan() with typed changed-artifact, evidence, signal, skipped-route, action, plan, and report rows.
• Added route-scoped maintenance_scan_router API discovery plus a maintenance-scan-template CLI scaffold.
• Added a FlowGuard self-model and tests that catch missing model/test alignment, missing StructureMesh routing, skipped-route drops, and broad claims without owner-route evidence.
• Updated docs and AGENTS guidance so non-trivial FlowGuard-managed work can route model/code/test drift, stale evidence, split pressure, and skipped routes to the owning maintenance skill without turning the scan into a heavy universal gate.

v0.39.4 - 2026-06-01

FlowGuard guidance surface fold-down.

• Folded model-first satellite trigger detail into a compact handoff table.
• Replaced duplicate kernel reference protocol copies with satellite-owned handoff stubs.
• Moved long ModelMesh and Model-Test Alignment delegation prompts into lazy-loaded template files.
• Added skill documentation tests and guidance-compression model checks to keep duplicate reference bodies and eager long templates from returning.

v0.39.3 - 2026-05-31

Codex skill diagram semantics sync.

• Added route-specific diagram intent guidance to installed FlowGuard skill surfaces.
• Clarified diagram edge meanings for development process, UI flow, model-test alignment, code-structure recommendation, and model mesh routes.
• Kept package runtime behavior unchanged.

v0.39.2 - 2026-05-31

• Added TestResultReuseTicket so previous test results can be reused only with current command, source, tested artifact, dependency, environment, result fingerprint, and coverage-scope proof.
• Extended Model-Test Alignment and TestMesh to reject reused test evidence unless it has both a current reuse ticket and current proof artifact.
• Updated public templates, docs, OpenSpec artifacts, and FlowGuard skill guidance for model/test result reuse and background evidence boundaries.

v0.39.1 - 2026-05-30

• Refreshed FlowGuard source-only release metadata after confirming the publication target.
• Updated package, README, and project adoption version surfaces to 0.39.1.
• No runtime behavior, helper API, schema, or model semantics changed from v0.39.0.

v0.39.0 - 2026-05-30

• Added Plan Detailing Compiler so rough ideas, short plans, and AI-generated outlines can be reviewed as typed PlanDetail rows before downstream FlowGuard routes run.
• Added projection helpers from PlanDetail to PlanIntake, Workflow Step Contracts, DevelopmentProcessFlow, and AgentWorkflowRehearsal.
• Added python -m flowguard plan-detailing-template, executable examples, public docs, API exports, and the flowguard-plan-detailing-compiler satellite skill.

v0.38.0 - 2026-05-29

• Removed compatibility-only public aliases that no current FlowGuard route should teach agents to use: FunctionResult.state and the old Plan Intake alias exports PlanIntakeSurface, PlanIntakeCompletenessFinding, FalseNegativeBackpropagationCase, FlowGuardClaimFinding, and review_plan_mutation_results.
• Added a legacy compatibility cleanup FlowGuard self-model that rejects done claims when agents delete safety classifiers or skip installed Codex skill content parity.
• Updated API/check-plan docs to describe core API semantic stability and route-first usage without preserving obsolete compatibility-only wording.
• Added tests that keep the removed aliases out of the public API while preserving SimilarityHandoff, route-scoped discovery, and compatibility-surface safety classifiers.

v0.37.0 - 2026-05-29

• Streamlined the AI-facing model-similarity surface: ordinary A/B/C maintenance reviews can now use model_signature_minimal(), model_signature_maintenance(), and model_similarity_plan_for_changed_member() before opening the full schema.
• Added SimilarityHandoff and ModelSimilarityReport.to_handoff() so downstream routes consume one typed handoff instead of repeated scalar relation/group/test/code obligation fields.
• Breaking cleanup: removed the old repeated similarity id fields from Existing Model Preflight, Code Structure Recommendation, Model-Test Alignment, and Architecture Reduction in favor of similarity_handoff.
• Updated route-first API docs and model-similarity templates so MODELING_HELPER_API is treated as the full index, not the first-read AI workflow.

v0.36.0 - 2026-05-29

• Split public template bodies into route-scoped internal modules while preserving the flowguard.templates facade and all template CLI commands.
• Added route-scoped API registry groups so callers can discover helper families without scanning only the flat MODELING_HELPER_API list.
• Added lightweight evidence gate/detail structures for gradual contraction of overloaded risk and development-process evidence fields while preserving skipped, stale, not-run, and progress-only visibility.
• Expanded Model Similarity Consolidation into a maintenance-oriented review: signatures can carry code/test paths, public behaviors, shared-kernel ids, adapters, maintenance tags, and changed refs; reports now derive maintenance groups, changed-sibling impacts, shared/variant test obligations, and code obligations without creating a separate audit route.
• Refactored public template/API tests toward table-driven coverage and added a FlowGuard structure-surface simplification model for template/API/evidence contraction hazards.

v0.35.0 - 2026-05-29

• Compressed FlowGuard AI-facing guidance so the kernel, satellite skills, and reusable AGENTS snippet now present a shorter first-read route: task triage, hard gates, the thin default model/check loop, compact route selection, and reference handoffs.
• Converted directly invokable FlowGuard satellite SKILL.md files into concise route shells while preserving standalone package verification, AGENTS/version adoption checks, evidence honesty, Mermaid snapshot guidance, and non-goals.
• Added guidance-compression FlowGuard checks that reject prompt-only completion claims until prompt budgets, validation, editable install, installed skills, shadow workspace, and local git evidence are aligned.
• Added prompt-budget tests for the kernel, satellite skills, and AGENTS snippet to prevent the hot path from regrowing into long protocol text.

v0.34.0 - 2026-05-29

• Added Model Similarity Consolidation so structured model signatures can be compared before agents create parallel model boundaries or shared code. The new route classifies same-workflow, family-variant, symmetric-flow, shared-kernel, duplicate-boundary, overlapping-ownership, parent/child, sibling-overlap, adapter-only, evidence-duplicate, false-friend, unrelated, and manual-review relations.
• Added ModelSignature, ModelSimilarityEvidence, ModelSimilarityRelation, ModelSimilarityPlan, ModelSimilarityReport, and review_model_similarity_consolidation() with evidence-gated recommendations and downstream handoffs to Existing Model Preflight, ModelMesh, Architecture Reduction, Code Structure Recommendation, StructureMesh, Model-Test Alignment, or manual review.
• Integrated model-similarity evidence into Existing Model Preflight, Architecture Reduction, Code Structure Recommendation, and Model-Test Alignment, plus CLI/template support through python -m flowguard model-similarity-template.
• Added Runtime Path Evidence so real code and tests can emit ordered runtime node observations that name their matching FlowGuard model ids, node ids, child model ids, proof artifacts, inputs, outputs, state, and behavior boundaries.
• Added RuntimeNodeContract, RuntimePathObservation, RuntimePathAlignmentPlan, RuntimePathAlignmentReport, RuntimePathRecorder, and review_runtime_path_alignment() with checks for missing, extra, stale, non-passing, out-of-order, model-binding, proof artifact, and boundary gaps.
• Integrated runtime path evidence into Model-Test Alignment, layered boundary proof, hierarchical ModelMesh child reattachment, Runtime Gateway Adoption, Closure Contract, workflow step contracts, CLI templates, docs, and Codex skill guidance so agents can compare code execution traces with parent and child FlowGuard model structure.
• Added Architecture Reduction compatibility-surface classification for old aliases, public facades, boundary adapters, archive-only surfaces, negative legacy tests, prune candidates, and evidence-needed compatibility surfaces before contraction is reported as ready.
• Archived the compatibility-surface and runtime-path OpenSpec changes into the main specs. Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.33.0 - 2026-05-28

• Added AgentWorkflowRehearsal so AI agents can take a fresh current-machine skill inventory before non-trivial multi-skill work, then rehearse selected skills, skipped candidate skills, ordering, continue/rework gates, validation gaps, side effects, and final evidence claims before execution.
• Added SkillInventorySnapshot, SkillCapability, AgentWorkflowPlan, AgentWorkflowStep, SkippedSkill, AgentWorkflowRehearsalReport, and review_agent_workflow_rehearsal() with pass, needs-revision, scoped, and blocked outcomes.
• Added the flowguard-agent-workflow-rehearsal Codex satellite skill as a peer route beside the existing FlowGuard satellites, plus executable example scenarios for stale snapshots, skipped required skills, wrong order, weak validation, missing rework gates, overbroad claims, and trivial over-trigger.
• Archived the project-adoption and agent-workflow OpenSpec changes into the main specs. Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.32.0 - 2026-05-28

• Added project adoption/version gate helpers so target repositories can carry a managed FlowGuard AGENTS.md block, .flowguard/project.toml, and adoption logs that name the FlowGuard GitHub repository and current package/schema versions.
• Added project-audit, project-adopt, project-upgrade, and project-adoption-template CLI/template entry points.
• Updated kernel and satellite Skill guidance so real target-project FlowGuard use checks or records the project AGENTS/version adoption state instead of relying on unstated machine-local knowledge.
• Updated project integration docs, README, API docs, OpenSpec artifacts, and focused tests. Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.31.0 - 2026-05-28

• Added obligation-family parity helpers so related obligations must carry member-by-member mechanism evidence before a family-level confidence claim.
• Added ObligationFamily, ObligationFamilyMember, ObligationFamilyEvidence, FamilyBadCaseSeed, derive_same_class_bad_cases(), and review_obligation_family_parity() with provenance-aware coverage matrices.
• Added AnalogousDefectCandidate, review_analogous_defect_scan(), scan-radius constants, and disposition states so post-miss reviews ask where the same failure shape may recur before broad closure.
• Integrated family parity into Model-Test Alignment and the Risk Evidence Ledger, and added analogous-scan gates to the Risk Evidence Ledger, so wrong provenance, missing sibling coverage, stale/non-passing family evidence, open analogous candidates, and scoped gates cannot be silently promoted to full confidence.
• Updated OpenSpec, README, docs, skill guidance, and focused tests for the new generic FlowGuard gate. Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.30.0 - 2026-05-28

• Added executable Closure Contract review helpers so broad done, release, publish, and production-confidence claims can be classified as full, scoped, or blocked from current route evidence.
• Added FlowGuardClosureContractPlan, ClosureEvidenceReport, RuntimeTraceMapping, ArtifactInvalidation, ModelQualitySignal, SameClassMissClosure, RuntimeGatewayInventoryClosure, and review_flowguard_closure_contract().
• Added the public closure-contract-template CLI/template scaffold plus README, API, closure-contract docs, and archived OpenSpec coverage for the final claim boundary.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.29.0 - 2026-05-28

• Added Runtime Gateway Adoption helpers so projects can distinguish design-only, test-aligned, and runtime-gateway FlowGuard adoption.
• Added RuntimeStateSurface, RuntimeGatewayContract, RuntimeWriteObservation, RuntimeGatewayAdoptionPlan, RuntimeGatewayAdoptionReport, and review_runtime_gateway_adoption() to block runtime protection claims when critical state writes can still bypass declared gateways.
• Updated OpenSpec, README, API docs, Model-Test Alignment guidance, and the closure contract so code-boundary evidence does not get overclaimed as full runtime state-write protection.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.28.1 - 2026-05-27

• Added lifecycle disposition support to Architecture Reduction candidates so completed or historical reductions remain visible as evidence without being re-reported as active ready work.
• Refreshed the structure simplification and public facade maintenance boundary so completed CLI/template and export-surface reductions close cleanly while shadow-only .flowguard duplicate cleanup stays a local sync concern.
• Updated OpenSpec, API-surface docs, agent guidance, and focused tests for the completed-candidate queue semantics.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.28.0 - 2026-05-27

• Added workflow step contracts so staged process plans can declare required, produced, and invalidated receipts and map them into DevelopmentProcessFlow validation gates.
• Connected workflow step contracts to claim readiness so done/release confidence can reject missing or stale step receipts instead of relying on prose-only process order.
• Updated public docs, examples, OpenSpec artifacts, and tests for the new workflow-step contract helper path.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.27.0 - 2026-05-27

• Added the FlowGuard model maturation loop with ModelMaturationSignal, ModelMaturationPlan, ModelMaturationReport, and review_model_maturation_loop() so post-code, post-miss, model-test, mesh, code-boundary, and freshness signals become explicit model-upgrade or scoped-claim decisions.
• Updated README, API docs, modeling protocol, closure contract, Risk Evidence Ledger guidance, DevelopmentProcessFlow, Model-Test Alignment, ModelMesh, and installed skill-source guidance so later evidence can force model refinement before broad confidence is claimed.
• Added focused unit tests for current, upgrade-required, scoped, blocked, mesh-driven, and caller-specified model maturation decisions.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.26.0 - 2026-05-27

• Defined the FlowGuard closure contract as intrinsic to complete FlowGuard use: plan/risk intake, model ownership, same-class model-miss evidence, model-test/code alignment, mesh or boundary proof when required, evidence freshness, Risk Evidence Ledger, and typed claim-chain support must be current before full done/release/publish/production-confidence claims.
• Updated README, AGENTS snippet, model-first skill guidance, modeling protocol, OpenSpec artifacts, and local self-review docs so partial model or test evidence is reported as partial/scoped FlowGuard evidence rather than complete FlowGuard use.
• Added a local executable closure-contract model and focused skill-doc tests that reject optional-mode framing and block completion without closure gates.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.25.1 - 2026-05-27

• Added ProofArtifactRef and legacy path disposition helpers so strict FlowGuard confidence checks can reject declaration-only passed/current evidence rows.
• Updated layered boundary proof, Risk Evidence Ledger, recurring model-miss defect-family gates, Model-Test Alignment, and DevelopmentProcessFlow with require_proof_artifacts gates that require result paths, fingerprints, current route evidence, and matching obligation coverage.
• Updated docs, skills, and templates so full confidence routes consume proof-artifact-bound evidence and explicit old-path disposition before closure.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.25.0 - 2026-05-27

• Added adversarial scenario synthesis on the existing Scenario Sandbox and model-first runner path so Explorer counterexamples, dead branches, exception branches, repeated state visits, and risk semantics can become candidate challenge routes without creating a parallel testing workflow.
• Updated retry, deduplication, cache, and side-effect packs to reuse the bounded input-shape fallback builder while preserving generated scenarios as needs_human_review candidate evidence until a domain oracle is supplied.
• Updated Scenario Sandbox, helper docs, README, OpenSpec artifacts, and focused tests for model-derived challenge-route evidence.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.24.0 - 2026-05-27

• Added model-impact freshness helper APIs so FlowGuard framework upgrades classify existing .flowguard models before broad freshness claims.
• Added ModelFreshnessRecord, UpgradeImpact, ModelImpactAssessment, ModelReuseTicket, ModelRerunEvidence, ModelImpactFreshnessPlan, ModelImpactFreshnessReport, and review_model_impact_freshness().
• Added an executable .flowguard/model_impact_freshness_gate model that rejects missing classification, blind reuse of old evidence, and affected models accepted without current rerun evidence.
• Updated framework-upgrade, DevelopmentProcessFlow, modeling protocol, helper layer, README, API surface docs, OpenSpec artifacts, and focused tests.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.23.1 - 2026-05-27

• Added plan-intake and typed claim-chain helper APIs so broad FlowGuard confidence cannot be promoted from under-declared plan inputs, lossy evidence adapter mappings, unbackpropagated false negatives, or known-bad mutations that still pass.
• Added PlanIntakeRiskSurface, EvidenceAdapterMapping, FalseNegativeCase, PlanMutationCase, FlowGuardClaimDependency, their review plans/reports, and the corresponding public review helpers.
• Updated public docs, model-first skill guidance, API surface exports, and focused tests for the new plan/adapter/miss/mutation/claim boundary.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.23.0 - 2026-05-27

• Added automatic model/test split diagnostics with AutoSplitCandidate, AutoSplitPolicy, AutoSplitPlan, AutoSplitReport, and review_auto_mesh_splits() so oversized, incomplete, slow, broad, progress-only, or release-only direct evidence routes to ModelMesh or TestMesh before broad parent confidence.
• Connected automatic split evidence to DevelopmentProcessFlow so done, release, archive, and publish claims stay blocked or scoped until current ModelMesh/TestMesh split gates exist.
• Added model/test split gate fields to Risk Evidence Ledger rows so final confidence can consume or reject required split evidence at the FlowGuard layer.
• Updated public templates, README, API docs, ModelMesh/TestMesh docs, DevelopmentProcessFlow guidance, installed skill protocols, and OpenSpec artifacts so automatic split is an existing-route evidence gate rather than a new first-class skill.
• OpenSpec validation, focused auto-split route tests, ModelMesh/TestMesh tests, public template tests, skill-doc tests, model examples, and full unittest discovery passed locally.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.22.2 - 2026-05-27

• Added model-miss closure roles to Model-Test Alignment so repaired misses can require both observed-regression evidence and same-class generalized test evidence before closure.
• Added recurring defect-family gate helpers so repeated or high-risk same-class model misses must name a promoted family, model obligation, authority boundary, observed failure, generalized case, historical holdout, and current external proof before full closure.
• Connected required defect-family gate status to the Risk Evidence Ledger so final confidence is blocked, scoped, or full at the FlowGuard layer instead of depending on a downstream product reminder.
• Blocked observed-bug-only, wrong-target, internal-path-only, and overclaimed evidence from satisfying same-class model-miss closure.
• Updated model-miss examples, public templates, FlowGuard skill protocols, README, API notes, check-plan guidance, and TestMesh handoff rules so tests must generalize with the upgraded model.
• OpenSpec validation, FlowGuard model-miss example checks, focused route tests, full pytest, and unittest discovery passed locally.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.22.1 - 2026-05-22

• Clarified the public README positioning: FlowGuard is a universal AI-agent skill layer and lightweight Python toolkit, while Codex support is an integration path rather than the boundary of the project.
• Updated the English and Chinese README intro so the first screen presents FlowGuard as a general AI-agent skill.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.22.0 - 2026-05-22

• Added explicit model-test evidence roles for primary evidence, primary edge_path evidence, leaf matrix-cell evidence, supporting contract evidence, and integration smoke evidence.
• Model-Test Alignment now treats multiple current primary edge_path proofs for the same obligation as a child_model_split_required blocker instead of allowing a label-only downgrade.
• Strengthened leaf boundary matrices with optional input/state axes, Cartesian completeness checks, unexpected-cell checks, missing observed behavior checks, and external-contract assertion-scope checks.
• Extended TestMesh so parent gates can require current passing evidence for exact leaf matrix-cell ids.
• Extended Existing Model Preflight so full parent/child preflight records layered proof evidence, parent coverage, child disjointness, child reattachment, and leaf boundary-matrix status before claiming the existing boundary is understood.
• Updated FlowGuard skills, docs, templates, and focused tests for the stronger leaf-boundary evidence path.
• Added a thin default AI entry path across the FlowGuard kernel, AGENTS snippet, README, API surface docs, and product architecture docs: model the risky boundary as Input x State -> Set(Output x State), add one invariant or scenario, run checks, inspect counterexamples, and escalate only when a named risk requires it.
• Reframed advanced satellites, helper APIs, meshes, ledgers, and framework suites as escalation paths rather than default reading for every FlowGuard task.
• Updated the skill-kernel FlowGuard model and skill-doc tests to catch a buried default path, advanced-route over-triggering, public/internal surface mixing, and stale satellite-count topology.
• Corrected the local satellite topology rollout model from the obsolete seven-satellite wording to the current ten-satellite topology.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.21.0 - 2026-05-22

• Added layered boundary proof helpers: ParentCoverageItem, ChildProofContract, ChildReattachmentProof, LeafBoundaryMatrix, LeafBoundaryMatrixCell, LayeredBoundaryProofPlan, and review_layered_boundary_proof(...).
• Added a four-table parent confidence gate for parent coverage, child disjointness, child reattachment, and leaf Input x State -> Set(Output x State) boundary-matrix evidence.
• Added python -m flowguard layered-boundary-proof-template, public docs, and focused tests for green proof, coverage gaps, illegal overlap, stale reattachment, missing leaf cells, output overflow, progress-only evidence, and too-large leaf split-required cases.
• Updated the FlowGuard kernel, ModelMesh, Model-Test Alignment, TestMesh, Existing Model Preflight, DevelopmentProcessFlow, Code Structure Recommendation, StructureMesh, Architecture Reduction, and Model-Miss Review guidance so parent/child model confidence now has an explicit leaf-boundary proof chain.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.20.0 - 2026-05-21

• Added code-boundary conformance helpers: CodeBoundaryContract, CodeBoundaryObservation, CodeBoundaryConformanceReport, and review_code_boundary_conformance(...).
• Extended Model-Test Alignment so a plan can include runtime boundary observations and block green alignment when real code accepts forbidden inputs or emits undeclared outputs, error paths, state writes, or side effects.
• Updated the model-test-alignment template, public docs, AGENTS snippet, and FlowGuard satellite skills so finite model-backed code boundaries require current boundary evidence before code conformance is claimed.
• Added an executable FlowGuard rollout model and focused tests for accepted inputs, rejected input gates, extra outputs, extra side effects, stale or internal-path-only observations, and alignment integration.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.19.1 - 2026-05-21

• Added Risk Evidence Ledger public APIs: RiskEvidenceRow, RiskEvidenceProof, RiskEvidenceLedgerPlan, RiskEvidenceLedgerReport, and review_risk_evidence_ledger(...).
• Added risk-evidence-ledger-template, public docs, and an executable rollout model that catches internal-path-only and progress-only final confidence claims.
• Updated FlowGuard kernel and satellite skill guidance so final done, release, publish, and full-confidence claims must connect user risks to model obligations, optional public code contracts, and current proof evidence.
• Extended adoption logging with risk_evidence_summary so scoped or blocked confidence boundaries are recorded alongside commands, skipped steps, and findings.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.19.0 - 2026-05-21

• Added Architecture Reduction as a FlowGuard route for using existing models to find behavior-preserving code contraction opportunities before implementation.
• Added public helper APIs ObservableArchitectureContract, ArchitectureReductionCandidate, ArchitectureReductionPlan, and review_architecture_reduction(...).
• Added candidate types, proof statuses, target actions, public-entrypoint StructureMesh gates, and target-structure handoff checks so risky shrinkage stays visible instead of becoming a silent rewrite.
• Added the flowguard-architecture-reduction Codex skill and updated companion FlowGuard skills so development, existing-model preflight, structure, mesh, model-test, and UI routes know when to invoke it.
• Added OpenSpec artifacts, local FlowGuard route-safety model checks, API docs, README coverage, and focused regression tests for the new route.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.18.6 - 2026-05-20

• Added validation-failure triage to DevelopmentProcessFlow so failed, stale, oversized, progress-only, or parent/child-sensitive validation is classified before agents keep patching or claim done.
• Added handoff guidance from DevelopmentProcessFlow to ModelMesh, TestMesh, and Model-Test Alignment for model-too-thick, test-too-thick, model-test mismatch, stale-evidence, and parent/child reattachment cases.
• Added Existing Model Preflight as a FlowGuard companion route for grounding discussion, proposals, bug fixes, feature work, refactors, prompts, skills, UI, test, and process changes in current FlowGuard model ownership before a new boundary is proposed.
• Added public helper APIs ExistingModelPreflight, ModelContextHit, ExistingOwnershipSnapshot, DuplicateBoundaryRisk, and review_existing_model_preflight(...).
• Added python -m flowguard existing-model-preflight-template, a Codex skill, OpenSpec artifacts, local FlowGuard model checks, README/API docs, and route trigger coverage for the new preflight path.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.18.5 - 2026-05-20

• Refined FlowGuard diagram guidance so agents choose route-specific diagram semantics before drawing: behavior/state, development process, UI state, model-test coverage, code structure, or mesh.
• Clarified that FlowGuard diagrams are standalone FlowGuard guidance and do not require LogicGuard or a shared cross-family protocol.
• Kept diagrams explanatory only; executable checks, evidence freshness, and route-specific validation remain the confidence source.

v0.18.4 - 2026-05-20

• Strengthened user-facing FlowGuard model visibility: non-trivial FlowGuard work now defaults to a Mermaid model snapshot during the work once the route or model shape is stable enough to explain.
• Extended route-specific diagram guidance across all installed FlowGuard satellite skills, while preserving concise output for tiny, obvious, direct-command, formatting-only, or user-suppressed tasks.
• Standardized global FlowGuard routing so clear staged-development, UI, structure, test, mesh, alignment, and model-miss work can select direct satellite skills instead of treating the model-first kernel as a universal first stop.
• Added OpenSpec artifacts and a local FlowGuard prompt-behavior model for the visibility rollout; diagrams remain explanation aids and do not count as validation evidence.
• Schema remains 1.0; runtime dependencies and public Python APIs are unchanged.

v0.18.3 - 2026-05-19

• Added lightweight user-facing Mermaid diagram guidance to the FlowGuard skill kernel so non-trivial model value can be explained when prose alone would hide the states, branches, gates, evidence, or claim boundary.
• Added route-specific optional diagram guidance for UI Flow Structure, ModelMesh, and DevelopmentProcessFlow while keeping diagrams as explanation aids, not validation evidence.
• Added OpenSpec artifacts and used a local FlowGuard self-model for the prompt behavior so the rollout rejects overbroad mandatory diagrams, shallow diagram guidance, and missing selected-route coverage.
• Schema remains 1.0; runtime dependencies and public Python APIs are unchanged.

v0.18.2 - 2026-05-19

• Added UI implementation validation for implemented/runnable UI completion claims, aligning user-visible feature contracts, reviewed UI journey coverage, and browser, desktop, or manual click-through evidence.
• Added public helper APIs UIFeatureContract, UIImplementationValidation, UIImplementationJourneyRun, UIImplementationStepEvidence, UIImplementationBlindspot, UIImplementationValidationReport, and review_ui_implementation_validation(...).
• Updated the UI Flow Structure template so generated scaffolds now demonstrate feature contracts, implementation journey runs, model revision/freshness, pure UI actions, residual implementation blindspots, and known-bad missing implementation evidence.
• Updated the UI Flow Structure skill, Skill Kernel route guidance, AGENTS snippet, API docs, UI docs, README, OpenSpec artifacts, and focused tests so "model-complete UI" is not confused with "running UI clicked through."
• Broadened DevelopmentProcessFlow routing guidance so staged implementation, validation freshness, archive readiness, and release confidence checks are selected earlier for non-trivial development work.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.18.1 - 2026-05-19

• Added ModelMesh mesh-closure meta-models so parent/child model handoffs can be checked as executable FlowGuard-style obligations before whole-flow parent confidence is claimed.
• Added public helper APIs MeshClosureModel, MeshClosureTransition, MeshClosureJoin, MeshClosureTerminal, MeshClosureReport, MeshClosureFinding, and review_mesh_closure_model(...).
• Updated review_hierarchical_mesh(...) so a declared closure model must pass before mesh_green_can_continue; closure blockers include missing root entries, unknown output references, unconsumed child outputs, incomplete joins, terminal leaks, missing out-of-scope rationale, and loop-like handoffs without progress evidence.
• Updated OpenSpec artifacts, ModelMesh docs, API docs, README, hierarchical examples, skill guidance, and focused tests for the new whole-flow closure gate.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.18.0 - 2026-05-19

• Added app-level UI journey coverage for complete UI claims, including launch state, top-level entry points, feature journeys, success terminals, failure/recovery/cancel/exit handling, terminal action allowances, and residual blindspots.
• Added public helper APIs UIJourneyCoverage, UIJourneyEntryPoint, UIFeatureJourney, UITerminalActionAllowance, UIResidualBlindspot, UIJourneyCoverageReport, and review_ui_journey_coverage(...).
• Updated the UI Flow Structure template so generated scaffolds now model launch, new-project, load-existing, failure, cancel, export, exit, terminal, and residual-blindspot coverage before structure and text hierarchy handoff.
• Updated the UI Flow Structure skill, Skill Kernel route guidance, AGENTS snippet, API docs, UI protocol docs, README, OpenSpec artifacts, and focused tests so "complete app UI" is not claimed from layout/text evidence alone.
• Added known-bad coverage cases for missing launch entries, unreachable path states, unknown path events, missing success terminals, unhandled failures, visible controls without modeled events, reachable events outside all journeys, terminal forward actions, misclassified terminal exports, and blindspots without validation boundaries.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.17.1 - 2026-05-19

• Added ModelMesh boundary-diff propagation so repaired child models can report unaffected, reattach_only, parent_rerun_required, sibling_rerun_required, or split_review_required before parent confidence is claimed.
• Added public helper API ChildBoundaryChangeSummary and summarize_child_boundary_change(...), and extended ChildModelEvidence with function, invariant, risk-class, and validation-evidence ownership fields.
• Updated ModelMesh review to expose boundary change decisions, reject point-fix-only bug-instance targets, and mark affected parent or sibling evidence stale when a child boundary changes.
• Hardened the public model-miss-template so generated review models now distinguish the observed issue, a same-class generalized bad case, and the known bug's holdout validation role before a repair can be finalized.
• Added generated negative scenarios that reject point-fix-only validation and validation that forgets to record the known bug as holdout evidence.
• Updated focused ModelMesh, public-template, Skill-doc, and OpenSpec artifacts for the boundary propagation and template hardening release.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.17.0 - 2026-05-18

• Added the ModelMesh child reattachment gate for local child model repairs: parent mesh confidence now requires current child evidence consumption plus stable input, output, state ownership, side-effect ownership, and outgoing contract handoffs.
• Added public helper API ChildReattachmentContract and extended ChildModelEvidence with evidence id, accepted inputs, emitted outputs, and incoming contract fields.
• Updated post-runtime model-miss review so a miss repaired inside a child model under a parent mesh remains open until the affected parent reattachment gate passes or records a blocker.
• Updated OpenSpec artifacts, Skill docs, public docs, examples, README, and focused tests to make "child-local green is not parent green" explicit.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.16.0 - 2026-05-18

• Added the UI Text Hierarchy Blueprint capability as the next public UI route. It reviews visible and assistive UI text by state, region, role, semantic key, owner, priority, duplication rationale, and warning/error escalation before copy, layout, or implementation flatten everything into equal prose.
• Added UI text hierarchy helper APIs: UITypographyToken, UITextElement, UITextHierarchyBlueprint, UITextHierarchyReport, and review_ui_text_hierarchy(...).
• Updated ui-flow-structure-template so generated UI models now run the full three-stage review: interaction model, structure derivation, and text hierarchy blueprint, including known-bad text hierarchy hazards.
• Positioned UI Text Hierarchy Blueprint as a sibling to UI Flow Structure: UI Flow Structure derives controls, regions, overlays, and stable placement from modeled interaction behavior, while UI Text Hierarchy Blueprint derives which headings, labels, helper text, status text, empty/loading/success/ failure text, CTA text, warnings, and errors should dominate or stay local in each state.
• Added OpenSpec artifacts under openspec/changes/add-ui-text-hierarchy/ covering proposal, design, tasks, and requirements for text inventory, primary/secondary hierarchy, duplicate semantic text, state-specific copy, blocking warnings, assistive rationale, and handoff boundaries.
• Updated README and product architecture docs to foreground the new capability while preserving the v0.15.0 UI Flow Structure material.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.15.0 - 2026-05-18

• Added the UI Flow Structure route for model-first interface design. The new helper first reviews a UI interaction model covering initial states, controls, events, state nodes, transitions, failures, recovery paths, terminal states, and control availability before deriving layout structure.
• Added UI flow structure helper APIs: UIControl, UIDisplayElement, UIStateNode, UITransition, UIInteractionModel, UIRegionRecommendation, UIStructureDerivation, UIFlowStructureFinding, UIInteractionModelReport, UIStructureDerivationReport, review_ui_interaction_model(...), and review_ui_structure_derivation(...).
• Added findings for missing initial states, missing availability matrices, unmodeled controls, failures without recovery, destructive primary/global controls, duplicate information in one state, same-level controls that trigger the same modeled function without a rationale, structure derivation before reviewed interaction evidence, missing parent surfaces, missing region maps, duplicate region ownership, duplicate information in one region, contextual controls placed globally, and overlays without origin or return paths.
• Added a ui-flow-structure-template CLI scaffold and public documentation for deriving first-level persistent areas, second-level contextual regions, third-level local actions, overlays, navigation ownership, and stable control placement from modeled UI behavior.
• Added the flowguard-ui-flow-structure Codex satellite skill and updated the Skill Kernel route map, AGENTS snippet, modeling protocol docs, API surface, product architecture docs, README, OpenSpec artifacts, and focused tests.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.14.0 - 2026-05-18

• Upgraded the Codex-facing skill architecture to one FlowGuard Skill Kernel plus seven directly invokable satellite skills: flowguard-model-test-alignment, flowguard-development-process-flow, flowguard-model-miss-review, flowguard-code-structure-recommendation, flowguard-model-mesh, flowguard-test-mesh, and flowguard-structure-mesh.
• Updated the global AGENTS snippet, README, product architecture docs, release checklist, OpenSpec artifacts, and skill-doc tests so the first-batch satellite topology is explicit while helper APIs and CLI templates remain package helpers rather than Codex skills.
• Added DevelopmentProcessFlow helper APIs: ProcessArtifact, ActionEffect, ProcessAction, ProcessEvidence, FreshnessRule, ValidationRequirement, DevelopmentProcessPlan, RevalidationRecommendation, ProcessFlowFinding, DevelopmentProcessFlowReport, review_development_process_flow(...), and derive_revalidation_plan(...).
• Added lifecycle findings for stale evidence after artifact changes, verifier changes after validation, model-test alignment evidence after model changes, requirement freshness propagation, unknown peer writes, ambiguous freshness policy, progress-only evidence, hidden skipped validation, failed/not-run evidence, missing V-style validation pairs, and release overclaims.
• Added a development-process-flow-template CLI scaffold and public docs for modeling development lifecycle ordering, artifact overwrite, evidence freshness, and minimum revalidation as a sibling route.
• Updated the model-first Skill Kernel, AGENTS snippet, API surface, README, OpenSpec artifacts, focused tests, and FlowGuard rollout model.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.13.0 - 2026-05-18

• Added conservative Python source-audit helpers for Model-Test Alignment: PythonCodeContractEvidence, PythonTestAssertionEvidence, ContractSourceAuditFinding, ContractSourceAuditReport, audit_python_code_contracts(...), audit_python_test_assertions(...), and review_python_contract_source_audit(...).
• The source audit checks real Python functions for declared symbols, external inputs, return values for external outputs, state writes, declared side effects, and side-effect-looking extra calls before trusting CodeContract rows.
• The test audit checks real Python tests for target code-contract calls and assertions before trusting TestEvidence rows that claim external-contract proof.
• Added source-level findings such as source_contract_missing_symbol, source_contract_missing_input, source_contract_missing_output, source_contract_missing_state_write, source_contract_extra_side_effect, source_test_missing_code_contract_call, source_test_missing_external_assertion, and source_test_internal_path_only.
• Updated templates, docs, Skill guidance, OpenSpec artifacts, focused tests, and a FlowGuard rollout model for source-audit hazards.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.12.0 - 2026-05-18

• Upgraded Model-Test Alignment so review_model_test_alignment(...) can compare FlowGuard model obligations, optional code external contracts, and ordinary test evidence in one direct review.
• Added CodeContract, code-contract role constants, test assertion-scope constants, and optional external-contract fields on ModelObligation, TestEvidence, and ModelTestAlignmentPlan.
• Added findings for missing code contract owners, code contracts that miss model-declared external behavior, exact contracts that add extra external behavior, missing code-contract test evidence, tests that bind only model obligations when code contracts are in scope, internal-path-only tests, unknown code contract references, duplicate code contract owners, and model-code-test binding mismatches.
• Kept model-test-only reviews backward compatible: code contracts are optional unless a plan explicitly requires them.
• Updated the Model-Test Alignment template, CLI help, public documentation, README, API surface docs, Skill Kernel route guidance, AGENTS snippet, OpenSpec artifacts, focused tests, and a FlowGuard rollout model for the new contract-alignment hazards.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.11.0 - 2026-05-17

• Added Code Structure Recommendation helper APIs: TargetModuleRecommendation, CodeStructureRecommendation, CodeStructureFinding, CodeStructureRecommendationReport, and review_code_structure_recommendation(...).
• Added review checks for missing source FlowGuard model evidence, missing parent boundary, missing target modules, missing FunctionBlock maps, missing validation plans, missing module rationales, unregistered owners, and duplicate FunctionBlock/state/side-effect/config ownership.
• Added a code-structure-recommendation-template CLI scaffold, public documentation, Skill Kernel route guidance, OpenSpec artifacts, focused tests, and a StructureMesh rollout model that catches target structures that are not model-derived or do not map model boundaries.
• Upgraded StructureMesh so existing large-script or large-module splits must include model-derived target code structure evidence inside the StructureMeshPlan; missing or mismatched target structure now blocks refactor confidence.
• Upgraded ModelMesh and TestMesh parent confidence checks to require FlowGuard-derived target split derivations before trusting child model or child suite/script ownership.
• Added ModelTargetSplitDerivation and TestTargetSplitDerivation helper records, protocol guidance, focused tests, and a FlowGuard rollout model for target split derivation coverage.
• Kept ordinary model-first work flexible: Code Structure Recommendation is a parallel route for direct architecture/file-split recommendations, not a hard gate for every FlowGuard model.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.10.0 - 2026-05-17

• Added standalone Model-Test Alignment helper APIs: ModelObligation, TestEvidence, ModelTestAlignmentPlan, ModelTestAlignmentFinding, ModelTestAlignmentReport, and review_model_test_alignment(...).
• Added coverage checks for missing model obligations, missing required test kinds, orphan or unknown test evidence, duplicate evidence ownership, stale/non-passing evidence, overclaimed confidence, and duplicate model obligation IDs.
• Added a model-test-alignment-template CLI scaffold, public documentation, Skill Kernel route guidance, OpenSpec artifacts, focused tests, and a FlowGuard rollout model that catches missing route visibility and accidental dependency on TestMesh, StructureMesh, or ModelMesh.
• Kept Model-Test Alignment independent from mesh split helpers: it compares explicit model obligations with explicit test evidence and does not split tests, refactor code, or read mesh reports.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.9.0 - 2026-05-17

• Added StructureMesh helper APIs for parent/child structure refactor governance: StructureMeshPlan, StructurePartitionItem, ModuleStructureEvidence, PublicEntrypointEvidence, StructureMeshFinding, StructureMeshReport, and review_structure_mesh(...).
• Added checks for missing or unregistered partition owners, duplicate partition owners, duplicate state/side-effect/config ownership, removed public entrypoints, missing facades, unsafe dependency cycles, config/default drift, missing or stale behavior parity, insufficient evidence tiers, and release-required parity gaps.
• Split routine and release refactor scopes so release-only structure evidence can remain visible as deferred obligations during routine work while still blocking release confidence when missing.
• Added a structure-mesh-template CLI scaffold, StructureMesh documentation, OpenSpec artifacts, Skill guidance, reusable AGENTS guidance, focused tests, and a FlowGuard rollout model for large script/module split governance.
• Modularized the model-first-function-flow Skill into a compact FlowGuard Skill Kernel plus dedicated sub-protocol references for core modeling, ModelMesh, TestMesh, StructureMesh, post-runtime model misses, conformance/adoption, long-check observability, and FlowGuard framework upgrades.
• Added a Skill Kernel rollout model that catches missing hard gates, route gaps, duplicate rule ownership, helper APIs mislabeled as sub-skills, standalone FlowGuard regressions, heavy-check over-triggering, and missing release/install sync routes.
• Clarified TestMesh as the test-side sibling of ModelMesh and StructureMesh: large test scripts, suites, or validation flows split into parent/child test hierarchy layers while parent gates consume child ownership and evidence contracts instead of expanded child internals.
• Added a soft oversize hint to the Skill Kernel so agents consider parent/child splits for large, slow, or hard-to-follow models, tests, scripts, modules, and commands without adding fixed thresholds or external planner dependencies.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.8.0 - 2026-05-17

• Added TestMesh helper APIs for layered validation governance: TestMeshPlan, TestPartitionItem, TestSuiteEvidence, TestMeshFinding, TestMeshReport, and review_test_mesh(...).
• Added parent/child test partition checks for missing owners, unregistered owners, duplicate partition ownership, duplicate state-write ownership, and duplicate side-effect ownership.
• Added evidence checks for stale results, hidden skipped tests, failed suites, timeout suites, insufficient evidence tiers, and background progress without final exit/result artifacts.
• Split routine and release validation scopes so release-only suites can remain visible as deferred obligations during routine work while still blocking release confidence when missing.
• Added a test-mesh-template CLI scaffold, TestMesh documentation, OpenSpec artifacts, Skill guidance, reusable AGENTS guidance, focused tests, and a FlowGuard rollout model that catches flat slow gates, missing/unregistered owners, duplicate owners, stale/hidden/timeout evidence, background progress-only overclaims, missing release evidence, and direct-test-runner scope creep.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.7.3 - 2026-05-16

• Added hierarchical model-mesh helper APIs for parent partition maps, child model evidence summaries, mesh findings, and mesh decisions.
• Added review_hierarchical_mesh(...) to check parent coverage gaps, unsafe sibling overlap, duplicate state-write ownership, duplicate side-effect ownership, stale/skipped child evidence, large-model split triggers, and legacy compatibility boundaries.
• Added classify_legacy_model(...) so existing model scripts can be registered and wrapped before they are trusted as strong child evidence.
• Extended model-mesh guidance to trigger from both model count and single-model scale, including estimated/observed state counts above 10,000, incomplete budgeted model groups, and models with unrelated functional areas.
• Added a nested hierarchy example, OpenSpec artifacts, focused tests, and a FlowGuard rollout model that catches missing scale triggers, coverage gaps, hidden overlap, duplicate ownership, legacy direct trust, child-graph expansion, background-check overclaims, and release-sync omissions.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.7.2 - 2026-05-15

• Simplified post-runtime model-miss review in the model-first-function-flow Skill to five practical miss types: boundary_missing, state_too_coarse, input_branch_missing, invariant_too_weak, and evidence_overclaimed.
• Required in-scope model misses to represent the observed issue plus one same-class generalized bad case when practical, so repairs do not stop at the exact bug that was just found.
• Kept the model-miss workflow lightweight: ordinary model misses do not add a default hazard registry, upgrade reviewer, model mesh, full coverage matrix, or evidence-level field.
• Updated OpenSpec artifacts, the reusable AGENTS snippet, modeling protocol, focused docs tests, and a FlowGuard rollout model for the revised workflow.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.7.1 - 2026-05-15

• Added a pre-implementation model-hardening gate to the model-first-function-flow Skill for complex optimizations, repeated bug repairs, stateful refactors, and model-miss-sensitive work.
• Required agents to write a change inventory, risk catalog, and risk-to-model coverage matrix before complex FlowGuard-backed edits.
• Clarified that representative known-bad hazards must fail before a model is trusted for the target bug class; happy-path checks alone are not enough.
• Added tiered handling for expensive project-specific model groups without hard-coding local model names as universally heavy or skippable.
• Updated the reusable AGENTS snippet, OpenSpec artifacts, focused tests, and a FlowGuard rollout model that catches code-first, happy-path-only, hard-coded heavy-model, peer-change, touched-heavy-skip, and premature-release variants.
• Ignored tmp/ so background check logs and template smoke outputs do not appear as release candidates.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.7.0 - 2026-05-14

• Added budgeted model-group execution for large reachable graph models via BudgetedGraphConfig and run_budgeted_graph_checks().
• Added a SQLite ledger that records seen, pending, processed, labels, failure samples, and shard summaries so repeated runs continue from pending work instead of starting over.
• Added fingerprinted run directories so changed model inputs, budgets, invariants, required labels, or caller-provided fingerprint parts do not reuse stale model evidence.
• Added whole-group reporting that distinguishes complete, incomplete, and failed; ok is true only when no pending states, failures, or missing required labels remain.
• Added shard-local progress plus model-group processed/pending totals on stderr, while preserving existing Explorer progress behavior.
• Added OpenSpec artifacts, a FlowGuard rollout self-model, focused tests, documentation, and a FlowPilot-style example for budgeted graph checks.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.6.1 - 2026-05-14

• Added a standard background-log contract for long-running FlowGuard checks: tmp/flowguard_background/ with stdout, stderr, combined output, exit-code, and metadata artifacts.
• Updated the model-first-function-flow Skill and reusable AGENTS snippet so agents must inspect actual log and exit artifacts before reporting long checks as complete.
• Clarified that direct Explorer(...) progress and legacy/custom runner final reports are different evidence types; final summaries are not live progress.
• Added OpenSpec change artifacts and doc tests that pin the log root, artifact names, completion evidence, and proof-reuse reporting expectations.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.6.0 - 2026-05-14

• Added an optional spec/SPAC-style skill-orchestrator collaboration model that keeps FlowGuard standalone while allowing upstream planning tools to hand off structured plans for risk review.
• Added docs/skill_orchestrator_collaboration.md with the three operating modes, the handoff contract, the upgrade sequence, known hazards, validation order, and non-goals.
• Updated the model-first-function-flow Skill, reusable AGENTS snippet, and project integration docs so upstream planner handoffs are optional context, not a FlowGuard dependency.
• Added tests that prove complete handoffs pass, missing upstream planners fall back to standalone FlowGuard, incomplete handoffs block collaboration only, and broken variants are caught for hard dependencies, hidden side effects, missing parallel ownership, skip-without-reason, ignored counterexamples, over-triggering trivial work, and completion without evidence.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.5.5 - 2026-05-13

• Added default ten-step progress output for Explorer.explore() so long serial model runs emit start and progress lines on stderr while preserving stdout for reports and JSON pipelines.
• Counted progress by top-level initial_state x input_sequence work units and kept the run serial and deterministic.
• Added progress_steps=0 and FLOWGUARD_PROGRESS=0 opt-outs for strict CI or callers that need silent runs.
• Added a FlowGuard rollout model plus focused tests for stderr routing, bounded output, small totals, opt-outs, runner inheritance, and report stability.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.5.4 - 2026-05-12

• Added Risk Purpose Headers to generated FlowGuard model templates so future agents can see the FlowGuard source repository, modeled workflow, guarded failure modes, use-before-editing guidance, and companion run command.
• Updated the model-first-function-flow Skill and reusable AGENTS snippet so AI-created or AI-updated FlowGuard model files carry the same lightweight header instead of only saying that they are FlowGuard artifacts.
• Used a local FlowGuard rollout model for this release that catches generic link-only headers, partial template coverage, missing Skill/AGENTS guidance, missing tests, manifest-style scope creep, and premature publication.
• Added focused tests for generated model template headers and Skill/AGENTS header guidance.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.5.3 - 2026-05-12

• Added a local model mesh protocol for projects with three or more FlowGuard models, including inventory fields, evidence tiers, required hazards, a prompt template, and a completion standard.
• Updated the model-first-function-flow Skill so multi-model FlowGuard projects must inventory and connect existing models before broad continue, release, completion, or production-confidence claims.
• Extended the FlowGuard Skill trigger self-review with a multi-model maintenance scenario and a broken variant that catches omitted model-mesh checks.
• Updated the reusable AGENTS snippet and modeling protocol docs with the model-of-models trigger.
• Schema remains 1.0; runtime dependencies and CLI templates are unchanged.

v0.5.2 - 2026-05-10

• Clarified the model-first-function-flow Skill around three broad FlowGuard scopes: Behavior Flow, Argument Flow, and Decision Flow.
• Added concise modeling hints for behavior state, reader/argument state, and decision/commitment state without adding new template families or changing the core API.
• Updated the README and reusable AGENTS snippet so users see that the existing model templates remain the execution scaffolds for all three flow types.
• Extended the Skill trigger self-review with structured argument and decision planning scenarios.
• Schema remains 1.0; runtime dependencies and CLI templates are unchanged.

v0.5.1 - 2026-05-09

• Refreshed the public project template so package-generated starter files match the richer model-first Skill template with validation, rejection, repeated-input handling, and traceability checks.
• Added public-safe Risk Intent + CheckPlan and post-runtime model-miss review template scaffolds, including CLI writers for project-template, risk-intent-template, and model-miss-template.
• Added template execution tests and privacy-marker checks to keep public scaffolds neutral and free of local project details.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.5.0 - 2026-05-08

• Added FlowGuardFindingLedger, FlowGuardFindingLedgerEntry, and build_finding_ledger to flatten summary findings and skipped/not-run gaps into one coverage-first repair ledger.
• FlowGuardSummaryReport.finding_ledger now exposes the ledger, and to_dict() includes it as machine-readable output for agents.
• Updated the helper-runner self-review with a broken variant that catches point-rule patches made before a full finding ledger is built.
• Updated the model-first Skill, AGENTS snippet, check-plan docs, framework upgrade guidance, modeling protocol, and README to route FlowGuard or LiveFlowGuard self-upgrades through coverage-first triage.
• Schema remains 1.0; runtime dependencies remain Python standard library only.

v0.4.2 - 2026-05-07

• Added Post-Runtime Model-Miss Review guidance to the model-first Skill: runtime, test, replay, or manual-validation failures after a FlowGuard pass must reopen the model-first loop instead of becoming direct patch-and-finish work.
• Clarified that agents should classify why the prior model missed the issue, represent in-scope misses as scenarios, invariants, replays, representative traces, or explicit out-of-scope boundaries, then rerun checks before validating the repair.
• Updated the reusable AGENTS snippet, modeling protocol, project integration notes, README, and Skill doc tests for the new completion gate.
• No core API, schema, runtime dependency, or CLI behavior changed.

v0.4.1 - 2026-05-07

• Clarified that when no FlowGuard model exists yet, the AI agent should create one from the current plan or adapt the included model template instead of refusing the task.
• Reframed the model as a fit-for-risk, customer-purpose artifact rather than always a minimal script: it should capture the failure modes the customer wants to expose and grow as new risks appear.
• Updated the README, AGENTS snippet, modeling protocol, project integration notes, and Skill wording to make evolving model scripts part of the public onboarding path.
• No core API, schema, runtime dependency, or CLI behavior changed.

v0.4.0 - 2026-05-02

• Added RiskIntent for explicit pre-modeling briefs that name failure modes, protected harms, model-critical state, adversarial inputs, hard invariants, and blindspots.
• Extended RiskProfile with an optional risk_intent field and audit suggestions for missing or thin Risk Intent Briefs.
• Updated the model-first Skill, modeling protocol, check plan docs, API surface docs, and README to make Risk Intent Briefs part of the public FlowGuard workflow.
• Fixed README version-section ordering so release notes read newest to oldest: v0.4.0, v0.3.1, v0.3.0, then v0.2.0.

v0.3.1 - 2026-05-02

• Added opt-in Mermaid source exporters for representative traces, generic state graphs, and loop review graphs.
• Exposed trace_to_mermaid_text, graph_to_mermaid_text, loop_report_to_mermaid_text, and mermaid_code_block through the public FlowGuard reporting helper API.
• Updated the README with English and Chinese Mermaid examples showing how FlowGuard turns a risky request into a finite model, reachable traces, findings, and optional conformance replay.
• Added a runnable examples/mermaid_export_example.py script that prints a Markdown Mermaid code block.
• Documented that Mermaid output is copyable text source and remains off by default so routine reports stay concise.

v0.3.0 - 2026-04-30

• Expanded the model-first-function-flow Skill from coding/repository-only framing to coding, repository, and process-design work.
• Added process_preflight mode for non-code or mixed workflows that need validation, adjustment, observation, or loss-prevention review before action.
• Clarified that booking, purchase, publication handoff, operational runbook, data migration, support escalation, and multi-agent coordination flows can be modeled when they have meaningful state, side effects, external dependencies, rollback concerns, or irreversible cost.
• Preserved the skip boundary for trivial reversible tasks and clarified that non-code process models are risk-discovery preflights, not proof of real-world prices, availability, policies, or vendor behavior.
• Updated the README, AGENTS snippet, Skill documentation, and Skill doc tests for the broader process-preflight scope.

v0.2.1 - 2026-04-30

• Simplified the model-first-function-flow Skill adoption-note wording so agents leave a short plain-language record instead of treating adoption logging as a heavy field checklist.
• Clarified that the adoption CLI can help create the log entry, but it is not a substitute for a short human-readable note when the model found something important.
• Updated the reusable AGENTS snippet with the same lighter note guidance.
• Clarified that SCHEMA_VERSION / python -m flowguard schema-version reports the artifact schema version, not the GitHub/package release version.

v0.2.0 - 2026-04-30

• Added optional standard property factories, model quality audit, RiskProfile, FlowGuardCheckPlan, and run_model_first_checks() to make model-first checks easier for AI coding agents without changing the core Input x State -> Set(Output x State) model.
• Added deterministic counterexample minimization, scenario matrix generation, unified summary reporting, and lightweight domain packs for deduplication, cache, retry, and side-effect risks.
• Added helper surfaces for adoption evidence review, state-write inventory, low-friction adoption logging, and optional maintenance workflow scaffolding.
• Expanded public examples, benchmark/evidence documentation, self-review models, and tests while keeping runtime dependencies at Python standard library only.
• Clarified that warnings, skipped checks, and pass_with_gaps are confidence boundaries, not hard failures and not production conformance claims.

v0.1.3 - 2026-04-29

• Added model-fidelity calibration guidance to the model-first-function-flow Skill.
• Added a modeling protocol step that treats FlowGuard models as falsifiable simulators of real workflows, not ground truth.
• Clarified that conformance replay and replay adapters must not hide relevant production behavior.
• Updated the reusable AGENTS.md snippet to record model-fidelity gaps and calibration changes.
• Refreshed the README positioning around FlowGuard as an architecture simulator / finite-state workflow simulator, including the hero tagline and AI-agent install path.