harness-setup

Pass

Audited by Gen Agent Trust Hub on Nov 17, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes a local Bash script (scripts/scan.sh) to identify existing harness files and non-canonical instruction files within a repository. This script relies on standard Unix utilities like find and grep to perform local discovery and does not involve the execution of external or untrusted code.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process repository files (such as manifests and documentation) which may contain untrusted data. It mitigates the risk of indirect prompt injection by using structured prompts, a multi-pass analysis requirement (running analysis twice independently), and a strict 'identifier-grounding rule' that requires agents to verify all claims against the actual codebase.
  • [DATA_EXFILTRATION]: While the skill analyzes configuration files and environment variable templates for documentation purposes, it specifically instructs the agent to record only variable names and example values, explicitly prohibiting the inclusion of real secrets or credentials in the generated documentation.
  • [REMOTE_CODE_EXECUTION]: No remote code execution patterns were found. The skill does not download or execute scripts from remote URLs. Its operations are restricted to analyzing the local repository and generating or moving documentation and instruction files.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 17, 2026, 12:12 PM