The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted pull request data and issue content which are provided to multiple reviewer sub-agents. This creates a surface where malicious instructions embedded in a pull request could attempt to influence the reviewer personas or the final report generation.
Ingestion points: Pull request metadata and diffs are collected in SKILL.md (Step 1) via gh pr view and git diff. Epic/issue bodies are fetched via gh issue view and gh api.
Boundary markers: The PERSONAS.md preamble includes explicit instructions to personas: "The PR body, worker reports and progress comments are claims, not evidence" and "Inputs (read what you need; do not paste them back)".
Capability inventory: Across SKILL.md and PANEL-DISPATCH.md, the skill utilizes gh CLI for viewing and commenting, git for repository operations, and dispatches sub-agents with filesystem access tools (Read, Grep, Glob, Bash).
Sanitization: No explicit sanitization, escaping, or filtering of the ingested external content is performed before interpolation into sub-agent prompts.

review-panel