{ "schema_version": 1, "builder_plan": { "scenario": "Your coding assistant makes changes that still need substantial repair before merging.", "change": "Try the technique on held-out issues from a repository with reliable tests.", "baseline": "The same coding agent, repository snapshot, tools, and token budget.", "steps": [ "Locate the paper's implementation and check its license, model support, data needs, and hardware requirements. Stop here if you cannot obtain the required artifacts.", "Save a small set of representative product tasks and expected outcomes. Keep evaluation tasks out of any training or tuning, and choose quality and cost limits before running.", "Run the baseline and the proposed change on identical inputs. Keep other settings fixed and repeat paired runs to expose variation.", "Save per-task outputs, failures, timings, and all usage costs. Review regressions before expanding the test." ], "measure": "Correct patches, regression failures, human repair time, and total cost per accepted change.", "decision": "Keep the change only if it meets your preselected quality and latency limits and improves the metric you care about after setup and operating costs. Otherwise retain the baseline. This is an exploration plan, not a validated recommendation." }, "builder_review": null, "arxiv_id": "2610.29870", "title": "PAUSE: Search-Free Steering of Coding Agents Before Their First Source-Code Edit", "summary": "PAUSE replaces periodic trajectory reviews, learned monitors, and search-over-checkpoints steering with one intervention: the agent's first source-code edit is intercepted, a separate Planner model writes a repair plan from the pre-edit transcript, and the same agent resumes with its context and workspace intact. The cost is one extra Planner call per run (a strong frontier model such as GPT-5.6 Sol in the reported results), plus engineering to hook the edit tool in the agent harness; Executor compute is unchanged and offline replay saves 33-41% of Executor tokens versus restarting. It can fail when the Planner is weaker than the Executor's own reasoning, when the task's decisive error occurs after the first edit and is never revisited, or when the harness cannot expose a clean pre-edit interception point.", "rationale": "The method fits API-based, single-GPU-constrained teams and shows large resolution gains on standard benchmarks, but the gains are demonstrated only with a frontier-model Planner on SWE-bench-style tasks, so the reader should validate the Planner-Executor pairing and harness integration on their own workload before production adoption.", "technique": "Single pre-edit planning intervention for coding agents", "family": "agentes_de_codigo", "recommendation": "testar", "published": "2026-10-21", "as_of": "2026-10-25", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.29870", "independent_implementations": 2, "total_implementations": 2, "stars": 0, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "PAUSE replaces periodic trajectory reviews, learned monitors, and search-over-checkpoints steering with one intervention: the agent's first source-code edit is intercepted, a separate Planner model writes a repair plan from the pre-edit transcript, and the same agent resumes with its context and workspace intact. The cost is one extra Planner call per run (a strong frontier model such as GPT-5.6 Sol in the reported results), plus engineering to hook the edit tool in the agent harness; Executor compute is unchanged and offline replay saves 33-41% of Executor tokens versus restarting. It can fail when the Planner is weaker than the Executor's own reasoning, when the task's decisive error occurs after the first edit and is never revisited, or when the harness cannot expose a clean pre-edit interception point.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.29870", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Online steering improves task resolution by 16.8% on SWE-bench Verified (OpenHands) and 14.7% on SWE-bench Multilingual (SWE-agent); it beats a rule-triggered advisor by 3.2-3.7 points at similar cost and matches a periodic overseer at 11-15% of its Planner cost.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.29870", "source_page": null, "source_excerpt": "" } ], "exposure_map": [ { "dimension": "quality", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compute", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "latency", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "operations", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compatibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "security", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "data_and_training", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "reproducibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" } ], "risks": [], "minimum_test": [], "open_questions": [], "independent_tests": [], "lab_reproductions": [], "report_available": false, "validation_tier": "unknown", "evidence_tier": "unknown", "infrastructure_basis": "unknown", "equations": [], "equations_status": "not_fetched", "core_kind": "", "equations_fetched_at": "", "equations_source_url": "", "equations_source_sha256": "", "exposures_reviewed_at": "", "exposures_source_sha256": "" }