shape-idea
Pass
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, such as third-party package documentation, maintainer statements, and project constraints. This creates a surface where malicious instructions embedded in external resources could influence the agent's behavior during the specification process.
- Ingestion points: External third-party source code, documentation, maintainer statements, and the project's
PRODUCT.mdfile. - Boundary markers: Absent. The instructions do not define specific delimiters or instructions to ignore embedded commands within the external content.
- Capability inventory: The skill writes to the local file system (
docs/specs/,GLOSSARY.md,DESIGN.md), executes benchmarks and experiments in scratch directories, and triggers other skills capable of installation and prototyping. - Sanitization: The instructions mandate that any conclusion about a third-party tool must be confirmed against the project's own versions in a scratch copy before implementation, providing a manual verification step.
- [DYNAMIC_EXECUTION]: The skill coordinates with the
add-stack-contextskill, which is explicitly described as 'owning discovery, source acceptance, installation' for new technologies selected during the shaping process. This represents a capability where design decisions can lead to automated software installation. - [COMMAND_EXECUTION]: The instructions direct the agent to run experiments, benchmarks, and previews to verify technical decisions. While the skill specifies using a scratch directory outside the working tree to mitigate side effects, this still involves the execution of code associated with the project being analyzed.
Audit Metadata