UserLevel plugin ecosystem for Ouroboros.
Ouroboros core should provide stable OS primitives. Plugins and first-party programs compose those primitives into domain-specific workflows.
+-------------------------------------------------------------------+
| Installable UserLevel Programs |
| |
| autoresearch gsd-agentos target-capabilities github-pr-ops |
| openai-skills-superpowers openhands-agentos merge-assistant |
| openhands-agentos merge-assistant jira-sync slack-incident |
| swe-agent-harness release-coordinator customer-debugger ... |
+-------------------------------+-----------------------------------+
|
| plugin contract / declared scopes
v
+-------------------------------------------------------------------+
| First-party UserLevel Programs |
| |
| ooo auto ooo run ooo pm ooo review? ... |
+-------------------------------+-----------------------------------+
|
| stable OS primitives
v
+-------------------------------------------------------------------+
| Ouroboros Core / OS |
| |
| Seed Ledger State Runtime MCP |
| Provenance Safety Boundaries Progress/Status Handoff |
+-------------------------------------------------------------------+
This repository is the home for the UserLevel plugin contract and reference plugin packages. It is intentionally not a registry server yet.
The initial goal is a local-only contract MVP:
- Plugin manifest format
- Command namespace declaration
- Core capability declaration
- External permission declaration
- Provenance and audit requirements
- Example UserLevel plugin packages, including external capability-assimilation adapters
- Anthropic Agent Skills assimilation reference plugin for
anthropics/skills - OpenAI/Agent Skills assimilation reference plugin for
openai/skills - External repository capability assimilation reference package
Ouroboros core owns primitives:
- Seed generation and validation
- Ledger and evidence/provenance tracking
- Durable workflow state
- Runtime/provider abstraction
- MCP tool surfaces
- Safety boundaries and permission checks
- Progress, recovery, and status reporting
- Execution handoff and result attachment
Plugins own domain-specific workflows:
- GitHub PR operations
- Autoresearch experiment handoff
- GSD AgentOS capability assimilation
- OpenAI/Agent Skill superpower assimilation
- SWE-agent issue-to-patch harness assimilation
- Merge assistance
- Jira/Linear synchronization
- Slack incident workflows
- Release coordination
- Customer debugging playbooks
ouroboros plugin add https://github.com/Q00/ouroboros-plugins --plugin autoresearch
ouroboros plugin add https://github.com/Q00/ouroboros-plugins --plugin github-pr-ops
ouroboros plugin add https://github.com/Q00/ouroboros-plugins --plugin gsd-agentos
ouroboros plugin add https://github.com/Q00/ouroboros-plugins --plugin openhands-agentos
ouroboros plugin add https://github.com/Q00/ouroboros-plugins --plugin swe-agent-harness
ouroboros plugin add https://github.com/<author>/<target>-ouroboros --plugin target-capabilities
ouroboros plugin trust github-pr-ops --scope github:read --scope github:pull_request:write
ooo auto-research prepare /path/to/autoresearch --goal "Improve validation bpb"
ooo gsd plan-phase 1
ooo gsd verify-work 1
ooo github-pr review https://github.com/org/repo/pull/123
ooo openhands agentos --workspace /path/to/repo --goal "Fix the failing test" --trusted-shell-execute
ooo superpower run openai-docs -- "explain how to build a plugin"
ooo swe-agent run --config config/default.yaml --problem_statement.github_url https://github.com/org/repo/issues/123
ooo github-pr merge --policy team-defaultOuroboros v0.39.1+ prompts for non-destructive required permissions during
plugin add, so autoresearch can grant filesystem:read and
filesystem:write during install. Destructive scopes, including PR write
scopes, still require an explicit plugin trust command.
For the full autoresearch workflow, including install, handoff preparation,
ouroboros auto execution, and the interview -> run -> evaluate -> reflect
boundary, see docs/autoresearch.md. A runnable
dependency-free experiment lives in
examples/autoresearch-char-lm.
The openai-skills-superpowers reference plugin shows how Ouroboros can assimilate the OpenAI/Agent Skills catalog without becoming a central marketplace. It projects skills into ooo superpower ... commands with manifest-backed permissions, audit/provenance events, trust gates, and handoff artifacts. See plugins/openai-skills-superpowers/README.md and docs/openai-skills-superpowers-pr-plan.md.
docs/
architecture.md Layer model and design principles
contract.md Plugin contract MVP
lifecycle.md Local-only plugin lifecycle
permissions.md Permission and trust model
audit.md Audit and provenance event model
migrations/ MAJOR-version schema migration guides (added when needed)
schemas/
0.1/
plugin.schema.json Draft JSON Schema for plugin manifests (v0.1)
audit-event.schema.json Draft JSON Schema for audit events (v0.1)
plugins/
anthropic-agent-skills/ Anthropic Agent Skills assimilation reference plugin
openai-skills-superpowers/ OpenAI/Agent Skills capability-assimilation adapter
autoresearch/ Autoresearch-to-ooo-auto handoff plugin
gsd-agentos/ GSD command-surface assimilation plugin
target-capabilities/ Reference external-repo assimilation package
opa-policy-gate/ OPA policy gate reference plugin
github-pr-ops/ Reference plugin skeleton
openhands-agentos/ OpenHands-to-AgentOS audited handoff plugin
catalog/
index.json Local catalog (boring index for reproducibility,
NOT a package-registry server)
A contract-validation GitHub Actions workflow runs the validator on every pull request and on pushes to main. To run it locally:
pip install -r requirements-dev.txt
python3 scripts/validate_contract.py
PYTHONPATH=plugins/autoresearch python3 -m ouroboros_autoresearch inspect /path/to/autoresearch
PYTHONPATH=plugins/autoresearch python3 -m ouroboros_autoresearch prepare /path/to/autoresearch --goal "Improve validation bpb"
PYTHONPATH=plugins/gsd-agentos python3 -m gsd_agentos validate-catalog
PYTHONPATH=plugins/gsd-agentos python3 -m gsd_agentos list
PYTHONPATH=plugins/github-pr-ops python3 -m github_pr_ops review https://github.com/Q00/ouroboros/pull/1
PYTHONPATH=plugins/openhands-agentos python3 -m openhands_agentos inspect
PYTHONPATH=plugins/target-capabilities python3 -m target_capabilities list-commands
PYTHONPATH=plugins/openai-skills-superpowers python3 -m ouroboros_superpowers catalog refresh --source-path tests/fixtures/openai-skills --ref fixture --out superpowers-catalog.json
PYTHONPATH=plugins/openai-skills-superpowers python3 -m ouroboros_superpowers --catalog superpowers-catalog.json run openai-docs -- "explain plugins"
PYTHONPATH=plugins/swe-agent-harness python3 -m swe_agent_harness inspect /path/to/swe-agent-outputDuring plugin development, install from the local checkout instead:
ouroboros plugin add . --plugin autoresearchThe first step is required: validate_contract.py imports jsonschema from
requirements-dev.txt. On a clean checkout, skipping it makes the validator
exit 2 with a "jsonschema is required" message.
Issue #45 is captured in docs/opa-policy-gate-pr-plan.md. The short version:
opa-policy-gate proves that a mature external policy engine can keep native
OPA/Rego workflows while gaining Ouroboros permission, audit, provenance, and
handoff semantics through the plugin contract.
Draft. This repository is for shaping the contract before committing to a full package registry or marketplace.
Issue #29 is captured in docs/assimilation.md, and the broader #27 authoring consensus is captured in docs/rfc/plugin-authoring-and-capability-assimilation.md. The short version: ouroboros-plugins is the contract/reference substrate, while an external target repository is assimilated as a permissioned, auditable, continuable plugin package. The target-capabilities package is the v0 reference fixture for that boundary.