The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's primary workflow involves executing shell commands to interact with GitHub (gh) and run project-specific validation tasks.
Evidence: Phase 1 instructions tell the agent to "Run the project's lint + type-check command. Check CLAUDE.md for the correct validation command".
[REMOTE_CODE_EXECUTION]: The skill allows cloning and processing remote repositories, which combined with the execution of commands found in those repositories, creates a remote code execution vector.
Evidence: Mode 2 allows users to provide a GitHub PR URL, which the skill clones to /tmp and then performs Phase 1 (automated command execution from the cloned content).
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes content from external repositories (PR descriptions, code, and project documentation) without sanitization or boundary markers.
Ingestion points: Project file CLAUDE.md, PR diffs, PR titles, descriptions, and author metadata.
Boundary markers: Absent. The skill passes untrusted PR content directly to parallel agents without explicit delimiters or instructions to ignore embedded commands.
Capability inventory: Shell command execution (via gh and CLAUDE.md strings), file system navigation, and network operations (via gh CLI).
Sanitization: None. The skill does not validate or sanitize the command string retrieved from CLAUDE.md before execution.

review-github-pr