The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's 'audit' and 'scan' modes ingest untrusted source code from a user-specified target path to perform security reviews. This creates a surface where malicious instructions embedded in the code being audited could attempt to influence the agent's behavior or findings.\n
Ingestion points: Source code files located in the TARGET_PATH or current project directory as described in SKILL.md.\n
Boundary markers: Absent; the agent is instructed to read code 'deeply' to find vulnerabilities without explicit delimiters or warnings to ignore instructions within the target code.\n
Capability inventory: The skill uses Bash, Read, Grep, and Glob tools (SKILL.md), and has the ability to spawn subagents for adversarial logic review.\n
Sanitization: No explicit sanitization, escaping, or filtering of the audited code content is mentioned before processing.\n- [COMMAND_EXECUTION]: The skill executes local shell scripts such as scanner-runner.sh and dast-runner.sh from its own library directory, as well as standard security CLI tools like grep, npm audit, and pip-audit. These actions are functionally necessary for security verification.\n- [EXTERNAL_DOWNLOADS]: The skill uses uv run --with pip-audit to dynamically fetch and execute the pip-audit utility. It also suggests using brew install nuclei for enhanced dynamic testing. These downloads target well-known and reputable security tools and package registries.

security-verify