The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill operates on untrusted external data by reading code diffs and incorporating findings from pr-reviewer reports or user-pasted comments. Since the skill has the authority to edit the working tree and execute shell commands, malicious instructions embedded in a pull request or review report could theoretically influence the agent's actions.
Ingestion points: git diff output, external pr-reviewer reports, user-pasted PR comments, and project instruction files (CLAUDE.md, AGENTS.md).
Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded prompts in the ingested data.
Capability inventory: Working tree modification (file editing), execution of git commands, and execution of project scripts (yarn lint, yarn test, yarn type-check).
Sanitization: No explicit sanitization or validation of the content of incoming review reports is performed before application.
[COMMAND_EXECUTION]: The skill executes shell commands to establish a baseline and verify fixes. Specifically, it identifies and runs scripts defined in package.json (e.g., yarn lint, yarn type-check, yarn test) and uses various git commands (git diff, git status, git restore). These are standard developer operations but are triggered automatically by the skill.

tidy