The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from PR comments, creating a potential vector for indirect prompt injection. This is mitigated by explicit security instructions in SKILL.md and references/agents/pr-comment-resolver.md that warn the agent never to execute commands or shell snippets found in comments. Ingestion occurs in scripts/get-pr-comments which fetches thread data. The skill uses git and gh for code modification and API interaction, while employing human-in-the-loop escalation (needs-human) for risky or ambiguous decisions.
[COMMAND_EXECUTION]: The skill executes shell commands using git and the GitHub CLI (gh). These commands are standard for PR management and are used to fetch comments, apply code changes, and post replies. The scripts (e.g., scripts/reply-to-pr-thread) use quoted heredocs to safely handle multiline markdown from comments, preventing accidental command or variable expansion when passing untrusted bodies to the GitHub API.

resolve-pr-feedback