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, andjqfrom 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 asagent-browser,jq, and the provideddownload-logs.shscript. These operations are consistent with the tool's intended functionality and do not involve untrusted remote execution.- [DATA_EXFILTRATION]: The skill reads theANTITHESIS_TENANTenvironment 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-browserand from downloaded JSON/text logs. - Boundary markers: Absent.
- Capability inventory: The skill performs subprocess calls to
agent-browser,jq,python3, andbash, and writes log files to local storage. - Sanitization: The
process-logs.pyscript strips ANSI escape codes, and theantithesis-triage.jsruntime cleans DOM text before it is extracted by the agent.
Audit Metadata