The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external codebases (mods), which is an inherent surface for indirect prompt injection. However, the skill implements exceptional defensive measures to mitigate this risk.
Ingestion points: Steps 1 and 2 involve reading arbitrary source code files from a provided codebase root.
Boundary markers: The audit prompts in references/audit-prompts.md use structured sections and clear constraints to separate instructions from the data being audited.
Capability inventory: The skill utilizes grep for reading and defines an editing flow for applying fixes in Step 5.
Sanitization: The 'Step 3 — Adversarial verification' phase is specifically designed to prevent the agent from acting on unreliable or poisoned data by requiring citation checking against the real file, an independent verifier pass with a fresh context, and a 20% self-sample gate.
[COMMAND_EXECUTION]: The skill uses deterministic mechanical checks via grep and powershell Grep to analyze codebase patterns. These are standard diagnostic tools used within the intended scope of a development audit and do not involve arbitrary or unsafe command construction.

rigorous-data-audit