The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection surface. The skill ingests untrusted data from GitHub pull request comments and reviews (including feedback from third-party AI bots) which is used to drive automated code modifications and git operations.
Ingestion points: gh api calls in rules/comment-fetching.md fetch content from PR reviews, line-level comments, and issue-level conversation.
Boundary markers: External data is encapsulated in a structured 'suggestion-pack' and quoted using markdown block quotes in templates/suggestion-pack.md before being handed to worker subagents.
Capability inventory: Worker subagents, as described in rules/handoff.md, have the authority to use Edit, Write, git commit, and git push tools.
Sanitization: The skill mitigates risks by requiring all actionable suggestions to pass through /critical (adversarial review) and /confidence validation gates.
[COMMAND_EXECUTION]: Utilizes shell commands via gh, git, and gw CLI tools for metadata resolution, worktree management, and repository synchronization.
[SAFE]: The skill implements defensive engineering practices, such as ensuring isolated worktrees per PR to prevent cross-contamination, refusing to perform force-pushes, and requiring a clean working state before proceeding.

implement-suggestion