pull-request

Fail

Audited by Gen Agent Trust Hub on Nov 9, 2026

Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONPRIVILEGE_ESCALATIONPERSISTENCE
Full Analysis
  • [PROMPT_INJECTION]: The skill instructions explicitly override standard agent safety and confirmation loops. It claims that invocation provides "explicit authority" for sensitive operations like add, commit, and push, and commands the agent to "never ask for a second commit/push confirmation" and "never ask the user whether to run tests or review."
  • [COMMAND_EXECUTION]: The skill executes several local JavaScript files using Node.js from the repository's .claude/ directory, including .claude/hooks/lib/doc-stamp-guard.cjs, .claude/scripts/lib/commit-review-policy.cjs, and .claude/hooks/lib/review-receipt.cjs. Running scripts found within the workspace being audited is a high-risk pattern.
  • [COMMAND_EXECUTION]: The protocol automatically executes "configured delta-scoped test/build/lint/contract commands" defined in project configuration files. This allows for arbitrary command execution controlled by the local project environment.
  • [DYNAMIC_EXECUTION]: The skill contains logic to dynamically load and execute other skills based on their name ($skill-name), enabling the execution of external skill definitions at runtime.
  • [PRIVILEGE_ESCALATION]: The skill grants itself the authority to perform potentially destructive Git operations, such as git push --force-with-lease, based on internal logic rather than explicit user approval for each specific instance.
  • [INDIRECT_PROMPT_INJECTION]: The skill defines a feedback loop where it ingests and processes untrusted data from external sources, specifically CI logs via gh run view --log-failed. This data is used to drive automated code repairs via the $fix skill.
  • Ingestion points: CI logs and PR status check outputs retrieved via the GitHub CLI.
  • Boundary markers: No explicit boundary markers or instructions to ignore embedded commands within logs are provided.
  • Capability inventory: The skill possesses extensive capabilities including file writes, Git history modification, and arbitrary command execution.
  • Sanitization: There is no mention of sanitizing or escaping log content before it is processed by the agent or repair skills.
  • [PERSISTENCE]: The skill includes instructions to run tasks in the background if they would exceed host tool timeouts ("run it in the background or re-run it"), which is a pattern used to maintain active processes outside the immediate user-visible session.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Nov 9, 2026, 01:34 AM