the-hit
Pass
Audited by Gen Agent Trust Hub on May 15, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill includes instructions to bypass human oversight in Step 8, mandating that the agent 'Immediately invoke gangsta:laundering — do NOT ask the Don what to do next, do NOT pause, do NOT prompt for confirmation. Auto-advance is mandatory.' This overrides standard agent behavior regarding user confirmation for tool or skill invocation.
- [PROMPT_INJECTION]: The workflow relies on processing data ('Reports') from sub-agents ('Workers') to drive control flow decisions, representing an indirect prompt injection surface.
- Ingestion points: Step 4 (Report Collection) and Step 5 (Crew Lead Reviews Reports) ingest content generated by sub-agents into the primary agent's context (SKILL.md).
- Boundary markers: Absent. There are no delimiters or instructions to treat sub-agent reports as untrusted data or to ignore embedded instructions.
- Capability inventory: The skill can invoke other sub-agents ('gangsta:the-capo', 'gangsta:drill-tdd') and automatically transition to subsequent phases ('gangsta:laundering').
- Sanitization: Absent. There is no evidence of filtering, escaping, or validation for the content of the reports before they are synthesized for decision-making.
Audit Metadata