subagent-driven-development
Pass
Audited by Gen Agent Trust Hub on Apr 14, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is a process-oriented orchestration framework for development and contains no malicious logic, obfuscation, or dangerous external references.
- [PROMPT_INJECTION]: The skill defines a pipeline for ingesting subagent artifacts. Ingestion points: Artifacts from subagents are received in SKILL.md (Phase 3). Boundary markers: Isolation is enforced through fresh context windows per task and compressed hand-off summaries (references/handoff-patterns.md). Capability inventory: The environment provides access to Bash, Read, Grep, and Glob tools. Sanitization: All ingested content undergoes mandatory two-stage reviews (references/review-rubric.md) which explicitly check for security and correctness.
- [DATA_EXFILTRATION]: The provided review rubric includes specific checks for data exposure, input validation, and injection prevention in the subagent's output.
- [COMMAND_EXECUTION]: Shell access is permitted for development tasks such as running tests, but the orchestrator is explicitly instructed to avoid direct implementation, which mitigates the risk of unauthorized or risky command execution within the orchestrator context.
Audit Metadata