The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external repository data and user-provided questions to generate opportunity portfolios and challenges.
Ingestion points: Consumes repo-context and task-question as primary data sources for its reasoning steps (e.g., in SKILL.md elicit and duel modes).
Boundary markers: The instructions explicitly require the agent to cite sources and separate observations from assumptions, creating a logical framework to distinguish source data from reasoning.
Capability inventory: Capabilities are restricted to local processing and writing JSON artifacts to .agents/scratch/ideas/. The skill utilizes shell execution for validation via local scripts.
Sanitization: Validation scripts (scripts/validate-challenge.sh and scripts/validate-output.sh) use jq to strictly enforce output schemas and data types, ensuring artifact integrity before handoff to downstream components like Plan.

idea-genie