All pull requests

Pull requests

• fix(codex): hand Codex the bare model slug and a codex that knows it (#1145)

  #1146·Benjamin-eecs opened Sep 25, 2026Contributor·

  ·
  1

  3

• fix(agents): bump ACP shims so claude-opus-5-5 and gpt-6 models can run

  #1139·alistairjcbrown opened Sep 23, 2026·

  ·

• feat(agents): add native OpenScience and DeepSeek ACP harnesses

  #1133·AzulEterno opened Sep 21, 2026·

  ·
  4

• fix(rollout): reclassify zero-activity bare timeouts; report measured elapsed

  #1131·AdnanElAssadi opened Sep 19, 2026·

  ·
  1

  2

• fix(review): record escaping workspace symlinks as evidence exclusions

  #1130·bingran-you opened Sep 15, 2026Collaborator·

  ·
  5

• fix(sandbox): upload denylist compose overlay on Daytona DinD

  #1123·bingran-you opened Sep 14, 2026Collaborator·

  ·
  1

• fix(metrics): make collect_metrics selection predicate shape-aware fo…

  #1121·VaisakhiMishra opened Sep 12, 2026Contributor·

  ·
  5

• fix(sandbox): reject pytest plugins backed by agent-writable code

  #1117·tulerfeng opened Sep 9, 2026·

  ·
  1

  1

• URL blocklist network mode enforced inside the sandbox (Daytona, Docker)

  #1115·zhiheng-yang opened Sep 9, 2026·

  ·
  6

• feat(sandbox): enforce agent egress blocklist with in-container proxy and firewall

  #1114·questiondlmarks opened Sep 9, 2026·

  ·
  7

• feat: add private research controls on top of URL denylist

  #1112·ElegantLin opened Sep 8, 2026Contributor·

  ·
  9

• fix: preserve safe ACP tool provenance in trajectories

  #1111·kywch opened Sep 8, 2026Collaborator·

  ·
  3

• feat: expose custom rollout planes through SDK.run

  #1110·kywch opened Sep 8, 2026Collaborator·

  ·
  3

• fix: disconnect agents before verification and harden cleanup

  #1109·kywch opened Sep 8, 2026Collaborator·

  ·
  5

• Post-hoc cost from recorded usage + OAuth-mode usage capture (bench cost, pricing table, OTEL receiver)

  #1108·yuchenlwu opened Sep 7, 2026Contributor·

  ·
  9

• Preserve Gemini passthrough trajectories in training exports

  #1106·kywch opened Sep 5, 2026Collaborator·

  ·
  3

• Fix Gemini CLI Vertex routing and authentication

  #1105·kywch opened Sep 5, 2026Collaborator·

  ·
  3

• fix(metrics): guard reward reads against a non-dict rewards field

  #1096·Galius5136 opened Sep 4, 2026·

  ·
  5

• fix(rollout): route verifier timeout cleanup by service

  #1095·zhiheng-yang opened Sep 4, 2026·

  ·
  2

• fix(rollout): clear stale LLM trajectory on rollout reuse

  #1094·zhiheng-yang opened Sep 3, 2026·

  ·

• refactor(agents): extract OpenClaw from BenchFlow core

  #1093·kywch opened Sep 3, 2026Collaborator·

  ·
  1

  15

• feat(review): integrate task rubrics into rollout scoring

  #1092·bingran-you opened Sep 3, 2026Collaborator·

  ·
  7

• fix(sandbox): verify the sandbox-user kill before scoring

  #1088·tulerfeng opened Sep 2, 2026·

  ·
  1

  3

• fix(rollout): skip process cleanup for Oracle

  #1085·zhiheng-yang opened Aug 31, 2026·

  ·

• feat(continue): typed continue-gate errors and batch skips instead of silent drops

  #1084·JeremyJC67 opened Aug 31, 2026·

  ·
  7