The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and review code from local repositories and pull requests, which is an ingestion point for untrusted data. This creates a surface for indirect prompt injection. However, the risk is mitigated by the skill's specific focus on static analysis and its use of well-defined reference policies that guide the AI to identify security flaws rather than execute instructions found in the code.
Ingestion points: Code files and Git diffs retrieved through the Read tool and Bash commands as described in SKILL.md and agents/code-reviewer.md.
Boundary markers: The skill utilizes comprehensive reference files (references/core-review-policy.md) to establish clear rules for review priorities and confidence filtering, helping to isolate the review logic from the content of the code being analyzed.
Capability inventory: Capabilities include reading files, executing Git commands via Bash, and interacting with version control forges (GitHub/Gitea) through the gh CLI or specialized skills.
Sanitization: The skill involves a manual verification step, as findings must be presented to the user for confirmation before any automated actions (like posting review comments) are taken.
[COMMAND_EXECUTION]: The skill uses Bash to execute standard development and version control commands, such as git diff, git status, and git rev-parse. These are necessary for the tool's primary purpose of analyzing code changes.
[EXTERNAL_DOWNLOADS]: The skill performs network probes to verify Gitea API versions and references documentation from established, well-known services and organizations, including Google and OWASP. These activities are transparent and consistent with the tool's auditing function.

pma-cr