dspy-simba
Pass
Audited by Gen Agent Trust Hub on Mar 22, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill implements a small-step optimization loop that analyzes training examples to generate "introspective rules" via a language model. This creates a surface for indirect prompt injection where malicious instructions embedded in training data could influence the generated logic.
- Ingestion points: Training examples are processed in the
optimizer.compile()method inSKILL.mdandexamples.md. - Boundary markers: The examples do not demonstrate the use of delimiters or specific instructions to the model to ignore embedded prompts within the training data.
- Capability inventory: The skill has the capability to write files to the local system using the
save()method as shown inSKILL.mdandexamples.md. - Sanitization: No sanitization of the training input is performed before the data is analyzed for rule generation.
- [EXTERNAL_DOWNLOADS]: The skill initiates network requests to OpenAI's well-known API endpoints to facilitate language model completions and optimization tasks.
- [COMMAND_EXECUTION]: The skill uses the
save()method to write optimized program configurations to the local file system in JSON format, which is a standard procedure for persisting model state.
Audit Metadata