The Agent Skills Directory

[PROMPT_INJECTION]: The skill ingests untrusted data from PR descriptions, commit messages, and source code diffs. This creates an attack surface for indirect prompt injection, where a malicious author could embed instructions in comments or metadata to trick the agent into performing unintended actions or applying harmful changes.\n- Ingestion points: PR bodies via gh pr view, commit logs, and the code diff.\n- Boundary markers: None explicitly used to separate untrusted content from agent instructions.\n- Capability inventory: The agent uses Bash, Edit, and Write tools, providing a path for injected instructions to modify the codebase.\n- Sanitization: No sanitization is performed on ingested data before it influences agent logic.\n- [COMMAND_EXECUTION]: The skill identifies and runs test suites and linters by executing commands detected in project files like package.json or Makefile. This constitutes execution of arbitrary code defined within the workspace.

review