The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in references/fix-loop.md require the agent to execute a specific Node.js script located at .claude/hooks/lib/review-receipt.cjs. The instructions include passing complex data types like JSON directly as command-line arguments (e.g., --snapshot-json='<exact saved JSON>'), which can lead to shell injection if the agent does not strictly escape the content.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from commit diffs and external project files.
Ingestion points: The skill reads commit candidates, project review rules, and affected source files.
Boundary markers: No specific delimiters or safety instructions are defined to separate the agent's instructions from the content of the files being reviewed.
Capability inventory: The skill has the capability to execute shell commands (node) and read the local file system.
Sanitization: There is no mention of sanitizing or validating the content of the commit or the project rules before they are processed or passed as arguments to the receipt script.

commit-review