dev-team

Pass

Audited by Gen Agent Trust Hub on Jun 16, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill uses external specifications ('verbatim spec') as instructions for sub-agents, which is a common surface for indirect injection.
    1. Ingestion points: Found in SKILL.md and all templates in references/prompt-templates.md.
    1. Boundary markers: No specific delimiters or safety warnings are used to wrap the untrusted spec data.
    1. Capability inventory: Agents possess shell access (Bash) to perform code edits, execute tests, and manage git commits.
    1. Sanitization: No input validation or filtering is performed on the provided specifications.
  • [DATA_EXPOSURE_AND_EXFILTRATION]: Explicitly enforces environment isolation by instructing agents to stay within a specific absolute workspace path and specifically prohibits touching the user's main repository checkout.
  • [SAFE]: Incorporates several security-best-practice constraints, such as mandatory verification for tests (mental-revert clause) and forbidding the use of git flags that bypass hooks or include untracked files.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 16, 2026, 08:15 AM
Security Audit — agent-trust-hub — dev-team