workflow-green-repo

Pass

Audited by Gen Agent Trust Hub on Oct 7, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon untrusted data from the repository, including source code, test logs, and build outputs. This creates a surface for indirect prompt injection where an adversary could embed malicious instructions within code comments or simulated error messages to subvert the agent's goal during the fixing phase.
  • Ingestion points: Reads repository source files, CI logs, and build artifacts during the 'Enumerate' and 'Fix root causes' phases.
  • Boundary markers: No specific delimiters or warnings are instructed to be used when the agent processes external command output or code content.
  • Capability inventory: The skill allows for command execution (via gate discovery), file reading, and file modification across the entire repository.
  • Sanitization: The instructions do not mention sanitizing or escaping content from external logs or files before the agent analyzes them.
  • [COMMAND_EXECUTION]: The skill requires the agent to discover and execute shell commands found within the target repository (e.g., in package.json, Makefile, or CI workflows). While these are typically standard developer tools like pnpm or make, the agent is instructed to run them automatically to establish a baseline and prove the 'green' status, which could lead to executing unexpected or malicious local scripts if the repository itself is compromised.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 7, 2026, 11:27 PM