The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several shell commands to perform its audit functions, including git clone --mirror for repository ingestion, bash to run scanning scripts, and the invocation of external security tools such as trufflehog and gitleaks.
[DYNAMIC_EXECUTION]: In Phase 2, the skill provides a template for subagents that involves dynamic construction and execution of shell commands. It references a "bundled sweep script" which is not provided in the skill definition, creating a dependency on external or hidden executable logic.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to process untrusted data from external repositories, which could contain malicious content designed to exploit the tools or the agent's logic during analysis.
Ingestion points: Client repositories (code, full git history, and binary artifacts) via mirroring.
Boundary markers: No explicit markers are used to isolate the data being scanned from the agent's execution instructions.
Capability inventory: The skill has access to shell execution, file system modification (cloning, writing reports), and network access (cloning and live verification).
Sanitization: The skill describes logic to identify false positives based on keywords (e.g., EXAMPLE, test) but lacks formal sanitization or sandboxing for the content being processed.
[DATA_EXFILTRATION]: Phase 3.4 describes a "live token verification" process. This involves transmitting discovered secrets (API keys/tokens) to external service providers' endpoints to check if they are active. While stated as an authorized activity, this involves sending sensitive data found in the environment to external third parties.

secret-audit