trace-analyzer

Pass

Audited by Gen Agent Trust Hub on Jun 22, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill is configured to execute specific git operations and a local Node.js script (.harness/scripts/harness-report.mjs). These actions are used to generate reports based on repository state and are limited to the local harness environment.\n- [PROMPT_INJECTION]: The skill processes untrusted telemetry and diff data, creating a potential surface for indirect prompt injection.\n
  • Ingestion points: Files under .harness/eval/results/, .harness/telemetry.jsonl, and the output of git diff.\n
  • Boundary markers: No explicit boundary markers or instructions to ignore embedded commands are present.\n
  • Capability inventory: The skill has access to Read, Grep, Glob, git, and local node script execution.\n
  • Sanitization: No explicit validation or filtering of the ingested telemetry content is defined.\n- [SAFE]: No evidence of data exfiltration, network requests to external domains, or hardcoded credentials was found. The skill operates exclusively on local files for failure classification and diagnostic recommendations.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 22, 2026, 05:40 PM
Security Audit — agent-trust-hub — trace-analyzer