The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from source code files and git diffs, which can contain malicious instructions hidden in comments or logic designed to manipulate the agent's output or actions.
Ingestion points: The skill ingests data via git diff output, user-provided file paths, and local repo documentation like AGENTS.md (Step 1).
Boundary markers: There are no explicit delimiters or instructions provided to the agent or sub-agents to ignore instructions embedded within the code content being reviewed.
Capability inventory: The agent has the capability to read the entire workspace, apply patches/edits to files (Step 4), and execute local shell commands for validation (Step 5).
Sanitization: The skill lacks sanitization or filtering of the ingested content before it is passed to the LLM sub-agents.
[COMMAND_EXECUTION]: The skill executes local shell commands to perform its core functions, including git operations and validation tasks.
Evidence: Step 1 uses git diff and git diff --cached. Step 5 executes 'smallest relevant validation' such as typechecks, compilers, formatters, or lint checks based on the project environment. These are legitimate uses for a developer-focused skill.

review-and-simplify-changes