dp-agent-eval-local
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The script
scripts/ask_local_deepworks.pydynamically modifies the Python system path (sys.path) based on a user-provided--repoargument and imports modules from that path at runtime. - [INDIRECT_PROMPT_INJECTION]: The skill takes user input from the
--messageand--working-dir-notearguments and interpolates them directly into the agent's prompt without sanitization or protective delimiters. - Ingestion points: CLI arguments in
scripts/ask_local_deepworks.py(line 74) which are interpolated into theDiagnosisCaseobject. - Boundary markers: None present; user input is concatenated directly into the prompt string.
- Capability inventory: The script utilizes an agent runner to interact with local workspace environments to perform diagnostic tasks.
- Sanitization: No escaping, filtering, or validation is applied to the input variables.
Audit Metadata