automated-search-over-agent-harness-mechanisms-gated-token-efficiency-only CLM-0234When an automated search over agent-harness mechanisms is gated on token efficiency with only a tolerance band on capability, the resulting full stack buys large token savings at a small but consistent score loss, and scores below the best single discovered mechanism — on EdgeBench the four-mechanism stack cut token traffic 44.7-49.0% and cost about a third while scoring 42.0 vs 44.8 for the base harness on GPT-5.6 Sol, whereas adding ObservationPack alone scored 47.2.
Capability: Improving its own scaffolding from failures
Observed on
Coding/terminal agent harnesses (Pi-based) on EdgeBench, Terminal-Bench 4 and IMO 2026, with GPT-5.6 Sol as search backend and Opus 5 as held-out backend; search optimized efficien.
Sources
- Measured on one held-out benchmark suite with single runs per configuration and no variance reported; Terminal-Bench 4 shows the stack solving fewer tasks (15 vs 18) at lower cost, consistent with the score-for-efficiency trade. Add-one ablations are on the same benchmark used for the headline result.
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- When an agent harness is evolved from execution traces under a fixed interaction budget, verifying each candidate edit on tasks chosen for the behavior it targets and accepting it only when the gain is attributable to that behavior beats scoring every candidate on a shared fixed, random or coreset task split, and larger verification budgets do not compensate: fixed-split methods using up to 24x the rollouts fell below the unmodified harness in half of their harness-benchmark pairs.Improving its own scaffolding from failures · reviewed by AI
- For LLM-driven heuristic discovery on combinatorial optimization tasks (TSP, CVRP, bi-objective and multi-component variants), letting an agent decide when to read, edit, diagnose and evaluate code while persisting each evaluation outcome as a retrievable experience card reaches better final heuristic quality than fixed evolutionary pipelines (EoH, ReEvo, FunSearch, MCTS-AHD, MEoH, MOTIF) at a fraction of their evaluation budget, and re-running the same task warm-started from the previous run's best code improves further round after round under a fixed budget.Fixing its own mistakes · reviewed by AI
- A large share of an agent benchmark score is attributable to the harness around the model rather than the model itself: holding the model fixed and changing only guides, sensors, retries and state management is reported to move the same model from 30.9% to 74.6% on GAIA and 30th to 5th on Terminal Bench — so a score reported for "a model" on an agentic benchmark is a score for a model-plus-harness, and comparisons across labs with different harnesses are not comparisons of models.Whether the measurement made the finding · reviewed by AI
- For CAE simulation tasks (OpenFOAM, FEniCS, PyChrono, COMSOL) run with a current general-purpose coding-agent harness and a strong backbone, a single agent with no multi-agent role decomposition and no scripted self-critique matches or exceeds the published scores of multi-agent specialized CAE systems, because the harness itself supplies the multi-turn reasoning and execution-feedback repair the decomposition was built to compensate for.Getting more from several models than from one · reviewed by AI
- Evolving an agent harness on 2,000 executable tasks disjoint from the evaluation benchmarks, by contrasting successful and failed trajectories on the same task to find recurring deficiencies and evolving five harness modules (agent loop, tool use, observation management, context management, completion detection) separately before integrating them, raised DeepSeek-V4-Flash from 47.6 to 52.4 on TerminalBench 2.0 and 73.4 to 76.5 on SWE-Bench Verified, transferred across domains and to other backbones frozen (GLM-5.2 59.6 to 61.8, MiniMax-2.5 41.6 to 44.9), while evolving the whole harness at once or without modules fell below the unevolved baseline (46.4 and 44.2).Improving its own scaffolding from failures · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: medium. Falsifier as drafted: A replication where the efficiency-gated full stack matches or exceeds the base harness score, or where the best single mechanism does not outscore the combined stack, on held-out tasks. Drafted stance toward a-fixed-model-can-improve-its-own-harness-with-a-regression-gated-loop: supports -- It is another regression-gated self-improving harness loop that improves a measured objective with the model frozen, here efficiency rather than pass rate.