pull-request

Fail

Audited by Gen Agent Trust Hub on Nov 10, 2026

Risk Level: HIGHPROMPT_INJECTIONDATA_EXFILTRATIONDYNAMIC_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill mandates the AI to bypass user confirmation for sensitive Git operations (add, commit, push, PR creation), asserting that the initial invocation provides absolute authority for all subsequent destructive actions.
  • [DATA_EXFILTRATION]: Instructions explicitly target sensitive file paths such as .env*, keys, and credentials, requiring the agent to list them in reports. This practice increases the risk of exposing sensitive configuration data or secret locations in the model's output.
  • [DYNAMIC_EXECUTION]: The workflow relies on executing multiple JavaScript files located within the repository (e.g., .claude/hooks/lib/doc-stamp-guard.cjs). This execution of local, unverified scripts presents a supply-chain risk if the repository content is untrusted.
  • [COMMAND_EXECUTION]: The skill makes extensive use of shell commands through git and gh to autonomously modify repository state and interact with remote GitHub services.
  • [INDIRECT_PROMPT_INJECTION]: The skill is instructed to ingest instructions from project files like CLAUDE.md and AGENTS.md, and to process CI logs without boundary markers or sanitization, creating a surface for external data to influence the agent's behavior.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Nov 10, 2026, 08:16 PM