The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses standard command-line tools such as git, find, npm audit, pip-audit, govulncheck, and cargo audit to analyze project structure, history, and dependencies. These operations are essential for its primary function as a security scanner.
[EXTERNAL_DOWNLOADS]: The skill invokes official dependency auditing tools (e.g., npm audit, pip-audit) which may connect to their respective public package registries to verify known vulnerability databases. These are trusted, well-known services used for standard maintenance.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data in the form of source code and pull request diffs, which represents a potential attack surface.
Ingestion points: Code changes and repository files are ingested via git diff and find commands as described in SKILL.md.
Boundary markers: The instructions do not define explicit delimiters to isolate code content from the agent's logic, though the context is clearly defined as a security review.
Capability inventory: The skill has access to shell execution via the agent's environment to run git and security tools.
Sanitization: No specific sanitization or filtering of the ingested code content is specified before the LLM processes it for vulnerabilities.

security-review