–
terminal-bench-loop
Runs a single Terminal-Bench task through a bounded, board-gated loop of smoke runs, diagnosis, and approved fixes until the task passes.
Lift · GLM-5.3-flash+25ppdirectional · all models →
87
Trustsecurity · quality · license · upkeep
Scan Saferisk 0/100
LicenseMITserved via MCP
HEAD:18f05fee5f80 Safe10/15/2026
Every source is polled for upstream changes (every 30 minutes for busy skills, adaptively less for quiet ones). A new commit becomes a pending version that must clear the same scan, judges and benchmark bar before it replaces this one — until then, agents keep getting the version you see pinned here.