The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of git diffs and project file contents, which are then used as input for five different analysis agents.
Ingestion points: The skill reads input from git diff, git diff --cached, and direct file reads via the Read and Glob tools in Step 2 and Step 3.
Boundary markers: The prompts for the sub-agents (e.g., Agents 1-5 in Step 3) interpolate the {diff} and {claude_md_contents} variables directly into the instruction strings without explicit delimiters or instructions for the sub-agents to ignore instructions embedded within the diff code.
Capability inventory: The skill possesses the Edit tool capability to automatically modify files (Step 5) and can execute shell commands via git and gh tools (Step 1, Step 3 Agents 3 and 4).
Sanitization: There is no evidence of sanitization, escaping, or filtering of the external code diffs before they are presented to the LLM-based sub-agents.
[COMMAND_EXECUTION]: The skill relies on executing various local development tools to perform its primary function.
Evidence: The workflow involves executing git diff, git remote show origin, git blame, git log, and various gh (GitHub CLI) commands including gh pr view, gh pr list, and gh api to retrieve repository metadata and history.

code-review