code-subagents

Pass

Audited by Gen Agent Trust Hub on Sep 18, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill facilitates a workflow where the output of 'implementer' agents is passed into the prompt context of 'reviewer' agents.
  • Ingestion points: The template references/batch-reviewer-prompt.md interpolates {BATCH_PATCHES} and {IMPLEMENTER_REPORTS}, while references/implementor-prompt.md interpolates {WORK_ITEM_REQUIREMENTS}.
  • Boundary markers: While the templates use Markdown headers (e.g., ## Task Patches), they lack robust delimiters or specific instructions for the agent to treat the interpolated data as potentially untrusted content.
  • Capability inventory: Implementer subagents are given file system write access and the ability to execute tests. Reviewer subagents analyze these results. A compromised or malicious subagent could embed instructions in its report or code comments to bypass reviewer scrutiny.
  • Sanitization: The skill does not define any sanitization or validation of subagent outputs before they are processed by the coordinator or reviewers.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 18, 2026, 05:54 AM
Security Audit — agent-trust-hub — code-subagents