Self-Harness runs an iterative loop of three stages: Weakness Mining, which extracts model-specific failure patterns from execution traces; Harness Proposal, which generates minimal harness edits tied to those failures; and Proposal Validation, which accepts an edit only after regression testing (paper). Instantiated from a minimal starting harness on Terminal Bench 2.0, SWE-bench Verified and AppWorld with MiniMax M2.5, Qwen3.5-35B-A3B and GLM-5, every final harness improved both held-in and held-out pass rates, with relative gains of up to 132%. In this corpus it is the system behind an agent rewriting its own harness, the item that gave the self-authored-scaffolding theme its name.
AISystem
Self-Harness
Shanghai AI Laboratory's paradigm in which an LLM agent improves its own operating harness through weakness mining, harness proposal and regression-tested validation, with no human engineers or stronger external agents.
Sources
Attribution and reporting linked by this record; links do not imply independent verification.
- Welcome to June 25, 2026 (theinnermostloop.substack.com)
Development timeline
Newest reporting dates first. Reporting dates may differ from event dates.
Cite or correct this record
Suggest a correction — opens a GitHub draft for you to review and submit.