The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external codebase content to identify shallow modules, which constitutes a vulnerability surface where malicious instructions could be embedded in code comments or documentation.\n- Ingestion points: The skill ingests file content from a user-defined repository, directory, or subsystem scope during Step 2 of the Audit mode process.\n- Boundary markers: The instructions require subagents to ground findings in specific 'named signals' and adhere to a strict 'Output Schema,' which provides structural constraints against accidental instruction override.\n- Capability inventory: The skill utilizes the 'multi-agent-analysis' tool to orchestrate analysis by subagents across different code clusters.\n- Sanitization: The skill does not explicitly detail sanitization, escaping, or filtering of the ingested code content before it is relayed to subagents for analysis.

deep-modules