shape-idea
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests project-level documentation and glossary files to ground its decision-making process.
- Ingestion points: Processes PRODUCT.md, GLOSSARY.md, and relevant files in docs/decisions/.
- Boundary markers: The skill employs a structured interview format, presenting candidates one at a time for user verification, which acts as a check against accidental instruction following.
- Capability inventory: Writing to docs/specs/ and the project glossary; invoking other agents/skills for prototyping and stack auditing.
- Sanitization: Relies on manual human review and explicit confirmation of every material branch before settling decisions.
- [COMMAND_EXECUTION]: The skill instructs the agent to run experiments, benchmarks, and previews in a scratch directory.
- Evidence: The instruction to run every experiment, benchmark, and preview in a scratch directory outside the working tree implies the use of local command execution for development tasks.
- Isolation: The skill explicitly mandates isolation from the main product source, configuration, and dependencies to prevent unreviewed changes.
- [EXTERNAL_DOWNLOADS]: The skill manages the discovery and installation of official and community skills for stack context.
- Evidence: Instructions include checking vendor control and requiring explicit approval before installation, adhering to safety protocols for third-party extensions.
Audit Metadata