Thank you for improving the Agile V Agent Skills Library. Contributions are licensed under CC BY-SA 4.0.
- Search existing issues and pull requests. Use the skill proposal form for a new skill.
- Keep changes concise, stage-focused, and consistent with
AGENTS.md. - Do not describe a draft skill as stable. A skill is draft only when its frontmatter contains
metadata.status: draft. - Do not claim certification, conformity, or guaranteed compliance. Describe evidence or alignment precisely.
Each skill belongs in a matching kebab-case directory with one SKILL.md. It must contain YAML frontmatter with name, description, license: CC-BY-SA-4.0, and quoted metadata.version, plus metadata.standard and metadata.author. The first heading must be # Instructions.
Use the repository ID conventions (REQ-XXXX, ART-XXXX, TC-XXXX), typed lineage, halt-on-ambiguity behavior, independent verification, human gates, and decision rationale. Prefer cross-references over duplicated contracts. Add adapted_from attribution when applicable.
If a skill contract changes, update that skill's metadata.version. Do not edit the repository version in package.json; release automation owns it.
- Create a focused branch and make the smallest complete change.
- Run
python -m pytest tests -qwhen Python and the dependencies inrequirements-test.txtare available. Otherwise, state the manual validation performed. - Validate changed YAML frontmatter, templates, schemas, links, and examples.
- When AI materially influences an artifact, create or update
AI_RUN_MANIFEST.yamlusingtemplates/AI_RUN_MANIFEST.yaml, link it to the evidence bundle, and exclude hidden reasoning, secrets, keys, and unredacted proprietary prompts. - Use a Conventional Commit title, such as
feat(skills): add rust build agentorfix(compliance): correct gate wording. - Complete the pull request template and identify draft contracts, tests, evidence, attribution, and unresolved risks.
Maintainers may request changes or decline proposals that duplicate existing skills, lack traceability, weaken role separation, or make unsupported assurance claims.