A place for distributable Agent Skills. They are placed at the repository root so
they are easy to find on GitHub and can be distributed via the gh skill
extension or by manual copy.
FSL work is split by layer so an agent does not turn PM requirements into design artifacts by accident — but the layers are also connected by refinement, and verifying that connection (cross-layer alignment / traceability) is one of FSL's primary uses, not a tail-end advanced topic.
A .fsl spec replaces the prose spec/design doc you would write anyway — one
source is both what people read and what the verifier checks, so there is no second
document to drift out of sync. That is why broad coverage across
business/requirements/design layers is the default target, not a per-feature
verification-ROI decision: high-risk-first is a valid adoption order, not the end
state of coverage. The resulting corpus is a living single source of truth, continuously
re-verified as it changes (regression, drift, cross-layer change-impact via
refine), and doubles as onboarding context for humans and AI alike:
| Skill | Use for | Main deliverable |
|---|---|---|
fsl/ |
shared language syntax, verifier commands, repair protocol, JSON result interpretation | core FSL reference |
fsl-business/ |
business flows, As-Is/To-Be controls, policies, KPIs, goals | business spec |
fsl-requirements/ |
PM/PdM requirements, acceptance criteria, forbidden flows, NFR/SLA | requirements spec |
fsl-design/ |
engineering design, internal state/actions, refinement mapping, testgen/replay handoff | kernel spec + mapping |
fsl-design-review/ |
design review, variants, SOLID/LSP/OCP/substitutability judgment | contract-conformance report |
fsl-requirements-document/ |
generating a stakeholder-facing requirements document from an already-checked spec; editing only its non-normative slot | Markdown requirements document + drift check |
fsl-from-code/ |
reverse-engineering a design spec from existing code, anchored by conformance back to that code | kernel spec + conformance harness |
fsl-delivery/ |
end-to-end FSL delivery orchestration from planning through implementation conformance | lifecycle status and gated handoff |
The shared fsl/ reference also covers three cross-cutting dialects that sit
outside the business/requirements/design progression:
dbsystem(fsl-db) — schema migration / artifact read-write compatibility checks andfslc db checkfindings; SQL/Prisma importers, runtime observation, bounded preservation checks, finite feature flags, and external DB evidence schemas are covered as explicit boundaries.domain(fsl-domain) — Functional DDD aggregate/command/event modeling and async effect lifecycles viafslc domain check; scaffold generation, runtime replay, and the production exactly-once/queue-delivery boundary are covered as explicit limits.ai_componentand recursiveagentcomposition (fsl-ai) — tool-authority hard contracts and agent-graph structure viafslc ai check/replay, plus the external statistical/migration/drift evidence commands (fslc ai eval/regress/compare/drift/compat); LLM truth, groundedness, and statistical quality claims stay outside formal proof.
The role-specific skills delegate syntax and verifier details to fsl/. Use the
narrowest role skill for authoring, then load fsl/ when writing syntax or
repairing verifier output.
Cross-layer alignment is one of FSL's primary uses. When the value is in the
seams — a requirement provably honored by the design, a control preserved across an
As-Is→To-Be change, end-to-end traceability — reach for fsl-delivery/. It
orchestrates business ⊒ requirements ⊒ design ⊒ implementation and gates the seams
with chain / refine / implements. Use a single role skill when one layer is the
whole job; use fsl-delivery/ whenever keeping the layers aligned is the deliverable.
fsl/SKILL.md— shared workflow, result→next-move repair protocol, minimal syntax, and structural rulesfsl/reference.md— condensed full language reference card for kernel FSL, compose, refinement, business/requirements dialects, and NFRfsl-business/SKILL.md— business-layer procedure and guardrailsfsl-requirements/SKILL.md— requirements-layer procedure and PM guardrailsfsl-design/SKILL.md— design-layer procedure and refinement guardrailsfsl-design-review/SKILL.md— review procedure and design-principle interpretationfsl-requirements-document/SKILL.md— requirements-document generation/editing procedure, permitted/forbidden agent operations, and thefslc document generate/claims/checkworkflowfsl-from-code/SKILL.md— code→design extraction procedure, the formalization-memo question set, and the two-axis anti-hollow conformance gatefsl-delivery/SKILL.md— lifecycle orchestration across business, requirements, design, verification, and implementation conformance
Claude Code inside this repository: no extra work is needed, because there are
symbolic links to each skill under .claude/skills/.
Using it in another project: copy the relevant skills/fsl* directories into
the target project's .claude/skills/, or into the user-wide
~/.claude/skills/. If you use the gh skill extension, specify this directory
(skills/) as the distribution source.
The verifier fslc itself is required separately (pip install -e . at the
repository root; the only dependencies are lark and z3-solver).