Contributing¶
This repo is an opinionated system. Changes should preserve coherence and prompting compatibility.
Start here¶
- Read
specs/000-index.mdandspecs/004-change-process.md. - Review taxonomy/workflow in
specs/003-taxonomy-and-workflow.md.
High-signal rules¶
- Treat skill names as API: prefer stable names; if you rename, capture it in a decision record and update docs (
specs/002-skill-contract.md). - For non-trivial changes, update the relevant spec and usually add an ADR in
specs/decisions/. - Keep
SKILL.mdlean; put depth inreferences/(progressive disclosure). - Prefer cross-links between skills over duplicating content.
Tool subprojects¶
The tools/ directory contains bundled tool implementations (e.g., tools/archobs/, tools/intelligence/). Each tool is a self-contained project with its own package manifest, license, and tests.
When modifying a tool subproject:
- Run the tool's own tests:
pytest -q tools/<tool>/tests(Python) ornpm testintools/<tool>/(Node.js). - Keep the corresponding skill docs (
skills/<name>/references/) aligned with tool behavior. - Tool subprojects may have different licenses than the top-level repo (e.g., MIT for archobs). Preserve license files within the tool directory.
- See
specs/decisions/015-vendor-archobs-tool.mdfor the rationale behind this pattern.
Verification¶
- Validate changed skills:
python3 .system/skill-creator/scripts/quick_validate.py skills/<skill-folder> - Validate repo-level docs/workflow consistency:
python3 .system/skill-creator/scripts/check_repo_consistency.py - Keep
README.mdandPROMPTS.mdaligned with the workflow-stage grouping (Define/Standardize/Harden/Verify). - Keep
specs/skills-manifest.jsonaligned with each skill's frontmatter metadata (stage, tags) and maintain manifest-only fields (trigger, related, overhead). - Build docs locally:
python3 .system/docs/generate_mkdocs_nav.py > mkdocs.yml && pip install mkdocs-material && mkdocs build --strict(seespecs/quickstart.md).
Feedback (what helps most)¶
If you file an issue or request a change, include:
- The prompt you used (and the tool/model if relevant).
- What you expected vs what happened.
- Links or snippets to the specific
SKILL.md/ spec section that felt confusing or missing.