The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, potentially untrusted data in the form of code files and git diffs, which serves as a surface for indirect prompt injection.
Ingestion points: Step 1 in SKILL.md identifies files via git diff, git show, and direct file reads to determine the review scope.
Boundary markers: The instructions do not define explicit delimiters or instructions to the agents to ignore embedded instructions within the source code being reviewed.
Capability inventory: The skill possesses the ability to execute shell commands (git, gh), spawn additional agents, and modify files (apply each fix directly in Step 3).
Sanitization: There is no evidence of sanitization or filtering of the ingested source code content before it is passed to the review agents.
[COMMAND_EXECUTION]: The skill executes various commands using standard development tools to resolve the project state and diffs.
Evidence: Uses git diff, git fetch, git show, and gh pr view --json baseRefName --jq '.baseRefName' to determine which files to review.
[DYNAMIC_EXECUTION]: The skill dynamically spawns six specialized sub-agents at runtime to perform parallel reviews.
Evidence: Step 2 in SKILL.md instructs the assistant to emit six tool calls for agents using the opus model to analyze the code based on specific prompts.

simplify-code