workflow-skill-creator
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes historical user workflow data to generate new instructions, which serves as an ingestion point for potentially untrusted data. This vulnerability is mitigated by mandatory brainstorming (Phase 1) and explicit design approval (Phase 2), ensuring human oversight of the generated logic.
- Ingestion points: User interaction history and brainstorming responses.
- Boundary markers: Structured iterative questioning and a required design approval step.
- Capability inventory: File system writes and execution of generated Python scripts.
- Sanitization: Relies on user review of the implementation plan to verify accuracy and safety.
- [DYNAMIC_EXECUTION]: The skill generates and subsequently tests new Python scripts based on a provided CLI template. This dynamic code generation is the primary purpose of the skill and follows a standardized, reviewable format.
- Evidence: Phase 3 and 4 involve writing and running .py files using the uv run command.
- [COMMAND_EXECUTION]: The skill utilizes command-line tools to execute and validate the generated automation scripts.
- Evidence: Explicit instructions in the implementation phase to test skills using standard execution commands.
- [EXTERNAL_DOWNLOADS]: The provided script template and generated skills are designed to interact with external APIs via network requests.
- Evidence: The reference template implements an API client using the standard Python urllib library for data retrieval and submission.
Audit Metadata