skills/edwinhu/workflows/dev-delegate/Gen Agent Trust Hub

dev-delegate

Pass

Audited by Gen Agent Trust Hub on Jun 18, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill invokes local utility scripts to manage workflow state. Specifically, it executes task-brief.sh to generate task-specific handoff files and review-package.sh to create git diffs for code review. These executions use the ${CLAUDE_SKILL_DIR} environment variable to resolve paths within the project structure.
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface (Category 8) typical of autonomous developer tools.
  • Ingestion points: Data enters the agent context from .planning/PLAN.md and from system git diffs via generated handoff files (e.g., .planning/handoff/task-[N]-brief.md).
  • Boundary markers: The skill uses markdown headers to structure handoff data but lacks explicit delimiters or instructions for subagents to ignore potentially malicious content embedded in the ingested data.
  • Capability inventory: The delegated subagents (specifically workflows:dev-implementer) are granted high-privilege tools, including the ability to create/edit files and execute arbitrary Bash commands to perform implementation and testing.
  • Sanitization: There is no evidence of sanitization or validation of the content read from planning files or diffs before it is passed to subagents.
  • [COMMAND_EXECUTION]: The dev-implementer subagent is explicitly granted access to the Bash tool with 'full implementation access' to run tests and verify code. This represents a high-privilege execution environment intended for development tasks.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 18, 2026, 06:50 AM
Security Audit — agent-trust-hub — dev-delegate