{ "schema_version": 1, "builder_plan": { "scenario": "An assistant chooses the wrong API or sends arguments your product cannot use.", "change": "Apply the proposed tool-selection or argument-generation change in a sandbox.", "baseline": "Your current agent with the same tool schemas and task 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 completed actions, invalid calls, retries, latency, and cost per completed task.", "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": "2609.01736", "title": "Harness Engineering in LLM Tool Use via Agent-Native Reusable Tool Primitives", "summary": "The method replaces rigid JSON-schema tool invocation and full in-context tool catalogues with LLM-wrapped tool primitives that accept natural language, plus retrieval of relevant tools from a central repository and a Planner-Router-Verifier orchestration loop. Costs include additional LLM calls per tool wrapper and per verification step, the engineering effort to wrap and index the tool catalogue, and dependence on the ToolFace repository being available or replicable. It can fail when the retriever surfaces the wrong tools at scale, when natural-language schema resolution misinterprets parameters, and when the claimed gains, which are self-reported against commercial baselines, do not transfer to the reader's specific tool set.", "rationale": "The pattern is implementable with API-only infrastructure and addresses real production pain points in tool calling, but the headline numbers are unverified against the reader's workload and require a pilot on their own tasks before adoption.", "technique": "Natural-language tool primitives with retrieval and planner-router-verifier harness", "family": "uso_de_ferramenta", "recommendation": "testar", "published": "2026-09-01", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2609.01736", "independent_implementations": 0, "total_implementations": 1, "stars": 5, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "The method replaces rigid JSON-schema tool invocation and full in-context tool catalogues with LLM-wrapped tool primitives that accept natural language, plus retrieval of relevant tools from a central repository and a Planner-Router-Verifier orchestration loop. Costs include additional LLM calls per tool wrapper and per verification step, the engineering effort to wrap and index the tool catalogue, and dependence on the ToolFace repository being available or replicable. It can fail when the retriever surfaces the wrong tools at scale, when natural-language schema resolution misinterprets parameters, and when the claimed gains, which are self-reported against commercial baselines, do not transfer to the reader's specific tool set.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.01736", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Reports ~10% average improvement over SFT-based models and ~6% over GPT-5.4, Claude-4.6-Sonnet, and Gemini-3.1-Pro across five benchmarks, up to 85% lower API cost, and 84% versus 22% average task completion (3.8x) on 50 real-world tasks.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.01736", "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_formula", "core_kind": "system", "equations_fetched_at": "2026-09-24", "equations_source_url": "", "equations_source_sha256": "", "exposures_reviewed_at": "", "exposures_source_sha256": "" }