The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute multiple shell commands (specifically ripgrep or grep) to perform its audit functions. These commands are used to search the local filesystem for hardcoded secrets, SQL injection patterns, and configuration errors.
Evidence: The 'Detection Patterns Quick Reference' section in SKILL.md provides specific commands such as rg '(password|secret|api_key|token)\s*=\s*"[^"]{8,}"' -i and rg 'shell=True' --type py to be executed by the agent.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and reason over untrusted external data (the user's codebase). This creates a vulnerability where instructions hidden in code comments or data files could influence the audit score or the agent's behavior.
Ingestion points: Ingests any project file detected during the 'Detect the Stack' phase, including source code (Python, Node.js, Go), configuration files (Docker, Kubernetes), and CI/CD manifests.
Boundary markers: The instructions do not mandate the use of XML-style delimiters or specific system-level warnings to distinguish between the skill's logic and the data being analyzed.
Capability inventory: The skill has read access to the project filesystem and utilizes internal scoring logic that could be subverted by adversarial text within the scanned files.
Sanitization: There is no evidence of input sanitization or filtering of the captured file content before it is processed by the agent's reasoning engine.

production-check