The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is explicitly designed to execute shell commands derived from the repository's command surface (e.g., scripts in package.json or Makefile targets). It utilizes a shared environment script at $SHARED_DIR/scripts/review_evidence.py to facilitate the selection and execution of these checks via a run-check protocol.
Evidence: The skill invokes python3 "$SHARED_DIR/scripts/review_evidence.py" select-checks and manages execution through a shared selector and transfer protocol.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface for indirect prompt injection because it parses and acts upon data found in untrusted repository files to determine its execution logic.
Ingestion points: The skill reads and interprets content from various repository files, including package.json, Makefile, Cargo.toml, go.mod, and CI configuration files (e.g., GitHub Actions or CircleCI configs) to discover gating commands.
Boundary markers: The skill does not implement explicit delimiters or "ignore instructions" warnings when processing the discovered commands from the repository configuration.
Capability inventory: The skill has the capability to execute arbitrary shell commands, read any file within the project directory, and write execution logs and evidence to a temporary directory outside the project.
Sanitization: There is no evidence of sanitization or validation of the specific command strings identified during the discovery phase before they are executed in the shell environment.

verify-changes