The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill coordinates a process where external code and delegate feedback are processed to automatically implement fixes, which presents a surface for indirect instructions to influence agent behavior. Ingestion points: The skill accepts implementation code from the user and receives feedback, scores, and recommendations from three distinct AI delegates. Boundary markers: There are no instructions to use delimiters or ignore embedded instructions when passing context to delegates or when interpreting their responses for subsequent code modifications. Capability inventory: The skill is authorized to perform file edits and invoke subagent tools through delegate_task based on the content of the external inputs. Sanitization: The workflow does not include a step to validate or sanitize recommendations from the delegates before the agent proceeds with applying changes to the project files.

council