The Agent Skills Directory

[SAFE]: The skill is restricted to drafting text and is explicitly prohibited from executing high-risk commands such as git commit, git push, or git add ., ensuring that all final actions remain under user control.
[SAFE]: It includes mandatory safety rules to prevent the inclusion of credentials, tokens, private URLs, or raw transcripts in the output drafts, protecting sensitive information from being recorded in version control history.
[SAFE]: The skill uses an 'evidence ledger' to strictly differentiate between observed and unobserved test results, preventing the generation of misleading verification claims in commit trailers.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an ingestion surface as it reads external data like pull request templates and git logs. However, the associated risk is extremely low because the skill lacks dangerous capabilities such as network access or arbitrary code execution, and it features instructions to process this data according to strict schemas.
Ingestion points: Reads .github/pull_request_template.md, git log, CONTRIBUTING.md, and AGENTS.md (specified in references/commit-and-pr-conventions.md).
Boundary markers: The skill uses defined sections and templates (e.g., repo_convention_read/v1, evidence_ledger/v1) to structure input data.
Capability inventory: Limited to text drafting and path listing. No subprocess execution or file system writes are authorized in SKILL.md.
Sanitization: Explicitly prohibits echoing raw transcripts or private data into the drafts.

omh-commit-pr-authoring