design-like-im-5
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: No malicious patterns, obfuscation, or high-risk behaviors were detected. All scripts are part of the skill's internal auditing and workflow management system. The use of subprocesses is restricted to executing internal Python scripts and the project's task runner (Mise).
- [INDIRECT_PROMPT_INJECTION]: The skill has a controlled surface for indirect prompt injection by processing user-provided intake data.
- Ingestion points: User-supplied intake facts enter the system through a JSON file passed to
scripts/run_pipeline.pyvia the--intakeargument. - Boundary markers: Intake fields are strictly constrained and validated against the
assets/run-intake.schema.jsonschema. - Capability inventory: The skill can write files to a designated run directory (via
scripts/lib/run_io.py) and execute internal audit scripts usingsubprocess.run(e.g., inscripts/build_examples.pyand the automated test suite inscripts/tests/). - Sanitization: Data undergoes structural and field-level validation in
scripts/lib/run_validation.pybefore work packets are generated for the model.
Audit Metadata