The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses several command-line tools to perform its auditing functions. It executes git for diffing branches and viewing commit logs, find for repository-wide file discovery, and various package security scanners including npm audit, pip-audit, govulncheck, and cargo audit. These tools are used for their intended purpose of identifying vulnerabilities in the target codebase.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data in the form of code changes, file contents, and dependency manifests. This creates a vulnerability surface where malicious instructions could be embedded in the code to influence the agent's reasoning or reporting.
Ingestion points: The agent reads untrusted content via git diff, git log, and file searches (find), as well as scanning source files and dependency lists (package.json, requirements.txt).
Boundary markers: While the instructions do not specify explicit delimiters for the scanned code in the bash scripts, the validation step (Step 6) includes logic to identify and exclude potential injection attempts or theoretical vulnerabilities.
Capability inventory: The skill can execute shell commands, read/write files within the .factory directory (which is a vendor-managed resource), post comments to pull requests, and create new git branches/PRs.
Sanitization: The skill employs a structured validation process that assesses exploitability and confidence scores, specifically filtering out user-controlled content in AI prompts to prevent accidental obedience to injected instructions.

security-review