Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note3 min read

Scrouting: Cost-Aware Routing of Coding Agents by Scouting the Repository First

Source

arXiv

arxiv.org/abs/2608.04804v1

Read the source

Source

Plain-Language Brief

Scrouting is a new system that aims to solve software issues more efficiently by first exploring a repository and then routing the issue to the appropriate model for resolution. This approach, presented in the paper 'Scrouting: Cost-Aware Routing of Coding Agents by Scouting the Repository First', uses a 7B searcher called SuperScout-7B to explore the repository and produce a structured handoff. The handoff is then used by a resume-based router to dispatch the task to one of four frontier fixers. This method has been shown to match the solve rate of the best single model at about a fifth of the total cost per solve, as reported on the full Python slice of SWE-bench Pro.

What the Document Says

  • The paper introduces SuperScout, which routes coding agents after scouting the repository first.
  • SuperScout-7B is a 7B searcher that explores the repository and produces a structured handoff.
  • The structured handoff is sandbox-verified and false claims are stripped before delivery.
  • A resume-based router dispatches the task to one of four frontier fixers based on the handoff and the task text.
  • Adding a new fixer requires no retraining.
  • On SWE-bench Pro, SuperScout matches the best single model's solve rate at about a fifth of the total cost per solve.
  • A no-router ablation, always the cheapest fixer with the handoff, ties the routed system on this benchmark.
  • The searcher's compute adds less than half a cent of GPU time per task.

Key Concepts Mentioned

  • Scrouting
  • SuperScout-7B
  • repository exploration
  • structured handoff
  • resume-based router
  • frontier fixers
  • SWE-bench Pro
  • cost per solve
  • GPU time

Claims Made by the Document

  • SuperScout matches the best single model's solve rate at about a fifth of the total cost per solve on SWE-bench Pro.
  • The handoff rather than the routing decision carries the result in the SWE-bench Pro benchmark.

Evidence Used

  • On the full Python slice of SWE-bench Pro (266 tasks) under the benchmark's official capped budget tier, SuperScout matches the best single model's solve rate (159 of 266 for SuperScout, 158 for the best model) at about a fifth of the total cost per solve.
  • A no-router ablation, always the cheapest fixer with the handoff, ties the routed system on this benchmark.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Frontier language models can resolve repository-level software issues, but each attempt is expensive, and existing routers select a model from the issue text alone. We present SuperScout, which routes after scouting the repository: a 7B searcher, SuperScout-7B, first explores the repository and produces a structured handoff whose reproduction claims are sandbox-verified, with false claims stripped before delivery. The searcher's hidden states, together with the task text, then feed a resume-based router that dispatches the task to one of four frontier fixers. Adding a new fixer requires no retraining. On the full Python slice of SWE-bench Pro (266 tasks) under the benchmark's official capped budget tier, SuperScout matches the best single model's solve rate (159 of 266 for SuperScout, 158 for the best model) at about a fifth of the total cost per solve, and the reported configuration sits above the random traffic-splitting baseline. A no-router ablation, always the cheapest fixer with the handoff, ties the routed system on this benchmark, so the handoff rather than the routing decision carries the result. A paired calibration study points to the mechanism: the handoff appears to redistribute rather than add solving ability, lifting the three cheaper fixers while slightly hurting the strongest, though at $N=99$ the per-fixer effects are directional only; the searcher's hidden states improve cost routing on the calibration labels while the handoff's own text does not. The searcher's compute adds less than half a cent of GPU time per task.