The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes the entire repository as untrusted data, allowing for potential manipulation of audit findings if files contain malicious instructions. Ingestion points: All repository files and documentation. Capability inventory: File read/write, shell execution (Bash), and subagent dispatch (Agent). Sanitization: Redaction of secret values is instructed, but no sanitization for prompt injection is performed. Boundary markers: No delimiters are used to isolate untrusted file content from the agent instructions.
[DATA_EXFILTRATION]: The skill systematically searches for and reads sensitive files such as .env, private keys, and cloud configuration. While it redacts values in the final report, the exposure of these file locations and their presence in internal artifacts creates a sensitive data surface.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to run various discovery and audit tools like find, grep, and package manager scanners, which interact with the project environment and its dependencies.

iso27001-audit