The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute complex search and manipulation commands across the local filesystem, including grep, git, and xargs to identify sensitive data.
[DATA_EXFILTRATION]: The skill targets highly sensitive information including AWS access keys, Stripe secret keys, and GitHub tokens. It specifically instructs the agent to read and scan environment files (.env) and JSON configuration files. Furthermore, it directs the agent to verify found secrets by making external network calls to provider APIs via WebSearch or CLI tools, creating a risk of exfiltrating discovered credentials to external endpoints.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted data from the local repository files being audited.
Ingestion points: Local source code, Git history, and configuration files are accessed via Grep, Glob, and Read tools as directed in the SKILL.md.
Boundary markers: No explicit delimiters or instructions to ignore embedded prompts are present when reading file contents.
Capability inventory: Access to Bash (shell execution), Write (file modification), and WebSearch (network access).
Sanitization: No sanitization or validation of the ingested file content is performed before processing.

secrets-audit