antithesis-triage

Pass

Audited by Gen Agent Trust Hub on May 8, 2026

Risk Level: SAFE
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill references dependencies such as snouty, agent-browser, and jq from trusted repositories belonging to Antithesis and Vercel Labs. These are well-known, official sources for the required triage tooling.- [COMMAND_EXECUTION]: Shell commands are used to invoke local utilities for report triage, such as agent-browser, jq, and the provided download-logs.sh script. These operations are consistent with the tool's intended functionality and do not involve untrusted remote execution.- [DATA_EXFILTRATION]: The skill reads the ANTITHESIS_TENANT environment variable and downloads logs from the vendor's domain to the local filesystem for analysis. No transmission of sensitive data to untrusted third parties was detected.- [PROMPT_INJECTION]: The skill does not contain any adversarial instructions or bypass attempts. Potential surfaces for indirect prompt injection are managed through the use of trusted data sources and local sanitization steps.
  • Ingestion points: Data enters the agent context from Antithesis report pages via agent-browser and from downloaded JSON/text logs.
  • Boundary markers: Absent.
  • Capability inventory: The skill performs subprocess calls to agent-browser, jq, python3, and bash, and writes log files to local storage.
  • Sanitization: The process-logs.py script strips ANSI escape codes, and the antithesis-triage.js runtime cleans DOM text before it is extracted by the agent.
Audit Metadata
Risk Level
SAFE
Analyzed
May 8, 2026, 12:12 PM