The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates an indirect prompt injection surface by aggregating data from external files and injecting it into the prompt of a downstream agent.
Ingestion points: The skill reads content from conductor/knowledge/patterns.md and conductor/knowledge/errors.json and processes user-controlled keywords from spec.md.
Boundary markers: There are no boundary markers or instructions within the generated 'Knowledge Brief' or the orchestrator example to prevent the downstream 'loop-planner' from executing instructions embedded in the retrieved data.
Capability inventory: The injected content influences the 'loop-planner' agent, which has the capability to define development tracks and generate code/project structure in plan.md.
Sanitization: No sanitization, validation, or escaping is performed on the data retrieved from the knowledge base before it is interpolated into the planner's context.

knowledge-manager