qoder-delegate

Pass

Audited by Gen Agent Trust Hub on Jul 31, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill's documentation (SKILL.md) provides a link to download the qodercli binary from its official documentation site at docs.qoder.com for installation and setup.
  • [COMMAND_EXECUTION]: The included scripts/relay.mjs script uses the node:child_process module to execute qodercli, git status, and process termination commands. The script is designed to run these tools without a command shell (shell: false) to prevent command injection vulnerabilities.
  • [PROMPT_INJECTION]: The skill facilitates an AI-to-AI delegation workflow which creates a surface for indirect prompt injection if the task briefs provided to the implementer are derived from untrusted data.
  • Ingestion points: Task briefs are read from files or stdin and passed as arguments to the relay script.
  • Boundary markers: The skill recommends using XML-like tags (e.g., <task>, <verification_loop>) within the briefs to structure instructions and separate task logic (references/writing-the-brief.md).
  • Capability inventory: The qodercli implementer has permissions to modify the project workspace, while the relay script manages the implementer's lifecycle.
  • Sanitization: No automated sanitization is performed on the brief; the skill mitigates injection risks by requiring the orchestrator to perform an independent diff review and run project gate commands before committing changes (references/review-and-land.md).
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 31, 2026, 05:32 PM
Security Audit — agent-trust-hub — qoder-delegate