The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Git commands (merge-base, diff, log, show) to inspect the repository. These commands are executed with flags such as --no-ext-diff and --no-textconv to prevent the execution of potentially malicious external tools defined in repository configuration.
[DYNAMIC_EXECUTION]: The skill executes a bundled Node.js script (scripts/check-identity.mjs) to verify the integrity and provenance of review reports. This script uses cryptographic SHA-256 hashing to detect stale evidence or unauthorized modifications to the analysis results.
[INDIRECT_PROMPT_INJECTION]: The skill inherently processes untrusted code and criteria as part of the review process. It mitigates injection risks through isolated execution contexts and strict evidence requirements. Findings are cited against committed source evidence and accepted criteria. Ingestion points: repository files and commit history. Boundary markers: instructions mandate fresh, separate read-only contexts for each reviewer axis. Capability inventory: read-only Git operations and local validation scripts. Sanitization: relies on structured JSON reporting and platform-level context isolation.

code-review