The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external feedback (human and automated) and repository templates, which are potential vectors for indirect prompt injection.
Ingestion points: human-review.md fetches feedback from repository tooling; pull-request-lifecycle.md reads repository templates and CI output.
Boundary markers: The instructions explicitly state "A comment is hypothesis, not authority" and require an "intake and critical analysis barrier."
Capability inventory: File editing, git commit, git push, and PR lifecycle management.
Sanitization: All feedback must pass through a multi-stage approval process, requiring explicit user alignment for each point and a consolidated plan before any implementation or publication occurs.
[COMMAND_EXECUTION]: The skill executes Git commands for branch management and synchronization.
Evidence: Uses standard, non-destructive commands like git status, git fetch, git switch, and git push in references/branch-and-sync.md.
Safety: The skill explicitly forbids git reset, git clean, git stash, and git push --force in SKILL.md and references/branch-and-sync.md.

git-workflow