The Agent Skills Directory

[SAFE]: The orchestration workflow for business and user analysis follows standard product discovery practices without malicious intent.\n- [SAFE]: Human-in-the-loop checkpoints (STOP — BLOCKING) prevent unauthorized operations or file modifications by requiring explicit user confirmation at critical stages.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from local files and user arguments. Ingestion points: docs/product/vision.md, docs/product/personas/, and $ARGUMENTS. Boundary markers: Absent. Capability inventory: Writing documents to docs/discovery/ and performing web searches. Sanitization: Absent. The workflow requires explicit user approval before identified opportunities or hypotheses are converted into files, effectively mitigating the risk of unintended actions.\n- [DATA_EXPOSURE_&_EXFILTRATION]: Web search is utilized for market research based on the generated hypothesis context. No sensitive system files, environment variables, or credentials are targeted, and the behavior is consistent with the skill's primary purpose.

recipe-discover