The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes build, test, lint, and typecheck commands (Gates) through the Bash tool. These commands are dynamically derived from the target project's configuration files (e.g., package.json, Cargo.toml). This is a standard and necessary function for a go-live audit.
[DATA_EXFILTRATION]: While the skill intentionally searches for sensitive information like committed secrets as part of its 'hygiene' audit dimension, it contains strict instructions and regular expression requirements to ensure all secret values are replaced with <REDACTED> before being reported, preventing accidental credential exposure.
[INDIRECT_PROMPT_INJECTION]: Because the skill is designed to read and analyze arbitrary codebase content from external repositories, it possesses an inherent attack surface for indirect prompt injection. Malicious instructions embedded in the analyzed code could attempt to influence the behavior of the 'Finder' or 'Verifier' agents.
Ingestion points: Uses Glob, Grep, and Read tools to ingest the entire target repository into the agent context.
Boundary markers: The skill defines a COMMON prompt block for agents but lacks strong boundary markers or explicit instructions to ignore embedded instructions in the audited code.
Capability inventory: The workflow utilizes Bash (limited to build/test gates) and the Agent tool to spawn up to 80 sub-agents.
Sanitization: The skill enforces strict redaction of credentials found in code to prevent them from entering the final report.
[REMOTE_CODE_EXECUTION]: The skill generates a JavaScript workflow script based on a bundled template and executes it using the Workflow tool. This is the primary mechanism of the skill and is controlled through a predefined control-flow structure.

go-live-audit