idea-to-blueprint
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a bundled Python script (
scripts/lint_blueprint.py) to validate the structural integrity and content requirements of the generated markdown blueprint. The script uses standard Python libraries and does not perform network operations or shell out to other processes. - [EXTERNAL_DOWNLOADS]: The research protocol (Phase 1) instructs the agent to perform extensive web research and fetch documentation from external sources to verify technology versions, licenses, and competitor features. This is performed using the agent's built-in search and fetch tools.
- [INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it ingests untrusted data from both the user (raw product ideas) and external web sources (market research and benchmarks).
- Ingestion points: Data enters the agent's context through user intake questions and web research fetching in
references/research-protocol.md. - Boundary markers: The instructions mandate strict evidence tagging (e.g.,
[VERIFIED],[ASSUMED]) to distinguish source data and require the agent to paraphrase rather than quote external content. - Capability inventory: The agent is authorized to write files (blueprints) and execute the bundled linter script.
- Sanitization: The linter script validates the presence of required IDs and sections but does not check for malicious instructions embedded in the content.
Audit Metadata