django
Pass
Audited by Gen Agent Trust Hub on Feb 18, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- Prompt Injection (LOW): The skill uses forceful constraints such as 'CRITICAL', 'MANDATORY COMPLIANCE', and 'NON-NEGOTIABLE' to prioritize its internal 6-step workflow over the agent's standard reasoning protocols.
- Indirect Prompt Injection (LOW): The skill establishes an attack surface by reading untrusted project data from external memory stores. 1. Ingestion points: Data is loaded in Step 2 ('memoryStore.getSkillMemory') and Step 3 ('contextProvider.getConditionalContext'). 2. Boundary markers: Absent; there are no instructions to the agent to delimit ingested data or ignore embedded directives. 3. Capability inventory: The skill is capable of writing output to the filesystem ('/claudedocs/') and updating persistent memory files via 'memoryStore.update'. 4. Sanitization: Absent; no logic is provided to validate or escape data retrieved from the project environment.
Audit Metadata