implement-dispatch
Pass
Audited by Gen Agent Trust Hub on Jul 30, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill processes output from worker subagents, which constitutes a surface for indirect prompt injection.
- Ingestion points: The orchestrator receives returns and synthesis reports from dispatched worker subagents (SKILL.md, "Dispatch cadence" Step 2 & 3).
- Boundary markers: Briefs composed for workers include explicit scope fences (ALLOWED/FORBIDDEN files and actions) and a divergence-escalation clause (SKILL.md, "Dispatch cadence" Step 1).
- Capability inventory: The orchestrator has the capability to perform file edits, git operations (status, add, commit, diff, log), and execute build/test commands (SKILL.md, "Dispatch cadence" Step 1 & 4).
- Sanitization: The skill mandates verifying all worker returns against direct evidence (such as diff reads, grep, and file reads) and running build/test commands on the main host rather than relying on worker reports (SKILL.md, "Dispatch cadence" Step 3 & 4).
- [COMMAND_EXECUTION]: The skill performs build and test commands on the host system to verify the work returned by subagents, ensuring that the main environment remains the source of truth for project integrity (SKILL.md, "Dispatch cadence" Step 4).
Audit Metadata