The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes application source code, which constitutes untrusted data ingestion. If the audited files contain malicious instructions, they could influence the agent's behavior.
Ingestion points: The skill reads all files within app/, config/, and other project directories using the Read, Grep, and Glob tools (SKILL.md).
Boundary markers: There are no explicit delimiters or instructions provided to the agent to disregard instructions potentially embedded within the code being audited.
Capability inventory: The skill utilizes Bash to run security scanners and test suites, alongside file system access tools.
Sanitization: No sanitization or filtering is applied to the content of the audited files before the agent processes them to generate a report.
[COMMAND_EXECUTION]: The skill executes local project binstubs and security tools, specifically bin/brakeman, bin/bundler-audit, and bundle exec rspec, to perform vulnerability scanning and policy verification.
[EXTERNAL_DOWNLOADS]: The skill performs an update of the vulnerability database using bin/bundler-audit check --update. This operation fetches data from the official RubySec vulnerability database.

security-audit