The Agent Skills Directory

[PROMPT_INJECTION]: Vulnerability to indirect prompt injection. The skill processes untrusted pull request data (code, comments, tests) which can contain malicious instructions intended to manipulate the agent's behavior.\n
Ingestion points: The review-pr command in references/commands/review-pr.md ingests data via git diff and gh pr view. Individual agents like comment-analyzer.md and code-reviewer.md also process file content.\n
Boundary markers: The system prompts for the agents do not include explicit instructions to disregard or isolate embedded commands within the code being reviewed.\n
Capability inventory: The skill has access to the Bash tool, file read operations, and the ability to spawn sub-tasks via the Task tool.\n
Sanitization: There is no evidence of input validation or sanitization for the code or comments being analyzed.\n- [PROMPT_INJECTION]: Deceptive metadata in SKILL.md. The skill claims to be authored by "Anthropic" and sourced from "Claude Plugins Official", which is inconsistent with the provided author context "ypyt1". This impersonation could mislead users into granting excessive trust to the skill.\n- [COMMAND_EXECUTION]: The skill uses the Bash tool to execute git diff and gh pr view commands to gather context about the pull request and its changes.

pr-review-toolkit