The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes git commits and diffs which are untrusted external inputs that could potentially contain malicious instructions to influence documentation drafting.\n
Ingestion points: Commit messages and file changes are read using git log, git show, and git diff as outlined in SKILL.md.\n
Boundary markers: The skill requires a mandatory human review and approval of the generated documentation diff and rationale before any external publication occurs.\n
Capability inventory: The skill can execute file writes, git commit, git push, and gh pr create via the scripts/doc-repo.sh utility.\n
Sanitization: The skill does not implement explicit sanitization or filtering of the content extracted from the ingested commits.\n- [EXTERNAL_DOWNLOADS]: The skill performs repository cloning from remote URLs provided in the user configuration file.\n
Evidence: git clone "$remote" "$doc_path" is executed within scripts/doc-repo.sh during the sync process.\n
Context: These operations are targeted at well-known services like GitHub and are fundamental to the skill's purpose of documentation management.\n- [DATA_EXFILTRATION]: The skill pushes local documentation changes to remote repositories and creates pull requests.\n
Evidence: git push and gh pr create operations are used in scripts/doc-repo.sh to publish approved changes.\n
Context: These network operations target authorized repositories and require explicit user approval for each specific contribution.\n- [COMMAND_EXECUTION]: The skill utilizes a local bash script to coordinate complex git and GitHub operations.\n
Evidence: scripts/doc-repo.sh is invoked with various commands such as sync, search, and publish.\n
Context: The script wraps standard development tools including git, jq, and the gh CLI, executing them with parameters derived from the agent's context and git state.

commit-documentor