The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from GitHub issue comments, creating an attack surface where the gate could be bypassed.
Ingestion points: The skill retrieves external data from issue comments using gh api "/repos/$REPO/issues/$ISSUE_NUMBER/comments" in SKILL.md.
Boundary markers: The logic relies on specific markdown markers like  and Review Status: COMPLETE to identify valid artifacts.
Capability inventory: The skill utilizes gh (GitHub CLI), git, and standard shell utilities (grep, jq) to perform its checks.
Sanitization: The skill uses jq and grep regex patterns to extract status and counts. While there is no direct command injection, the gate logic could be subverted if a malicious user posts a comment that perfectly mimics the expected machine-readable format to trick the agent into believing a review is complete.

review-gate