The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing external data that could contain malicious instructions.\n
Ingestion points: Reads and processes untrusted data from user-provided plans, git diffs, and project configuration/memory files such as CLAUDE.md, AGENTS.md, and package.json.\n
Boundary markers: The instructions do not define explicit delimiters or instructions to ignore potential commands embedded within the diffs or plans being reviewed.\n
Capability inventory: The skill can execute shell commands (git, npm, make, just) and perform file writes to update project rules and documentation.\n
Sanitization: There is no evidence of sanitization or filtering applied to the ingested content before it is processed by the agent.\n- [COMMAND_EXECUTION]: The skill interacts with the local system by executing several shell commands to facilitate its review process.\n
Execution details: It utilizes git status, git diff, and git remote show origin to establish the scope of code reviews. It also instructs the agent to discover and run verification commands such as npm test, npm run lint, or commands found in Makefile and justfile. While these are standard development tasks, they involve executing code and configurations defined within the user's project environment.

stage-gate