learning-capture
Pass
Audited by Gen Agent Trust Hub on May 3, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [SAFE]: No malicious patterns, hardcoded secrets, or unauthorized network operations were detected in the skill instructions or reference files.
- [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection as it processes data from an external database and user input.
- Ingestion points: Reads hypothesis, metrics, and experiment data from the database using the
get-projectcommand (SKILL.md). - Boundary markers: No delimiters or 'ignore' instructions are present to isolate ingested data from the prompt.
- Capability inventory: The skill has the ability to update database records and write files to the local file system (SKILL.md).
- Sanitization: No sanitization or validation of external data is specified before its use in generating reports or updating state.
Audit Metadata