The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze arbitrary local codebases, creating a surface where malicious instructions embedded in source code (e.g., comments or documentation) could attempt to influence the agent's synthesis or drive future exploration steps.\n- Ingestion points: The agent ingests data from the local filesystem using tools such as Read, Grep, Glob, and Bash.\n- Capability inventory: The skill possesses the Agent capability to spawn sub-processes and the Write capability to create and update local files like seed-questions.json and ARCHITECTURE-SUMMARY.md.\n- Sanitization: The skill mitigates the impact of potential injections by enforcing a strict 'read-only' boundary for mapper subagents, ensuring they cannot edit the codebase or access the web.\n- Boundary markers: The agent relies on internal logic and inherited guardrails from the loop-controller skill to delineate instructions from the data being analyzed.

codebase-exploration-loop