All pull requests

Pull requests

• fix(harness): capture agent exit status

  #484·mikemikimike opened Nov 5, 2026·

  ·
  1

• Allocate local viewer ports atomically through the container engine

  #429·Sinkleberg opened Nov 3, 2026·1 task done·

  ·
  1

• fix(ci): select all downstream harnesses when runtime-server changes

  #413·phatloi0010 opened Nov 2, 2026·

  ·
  1

• fix(runner): preflight extra_info staging against basename and personal-info collisions

  #412·phatloi0010 opened Nov 2, 2026·

  ·
  1

• fix(analyze): read lenient rescores and report overall two-stage success rate

  #411·phatloi0010 opened Nov 2, 2026·

  ·
  1

• Fix judge handling of truncated request bodies

  #410·mikemikimike opened Nov 2, 2026·

  ·
  1

• fix(runner): choose Chat Completions sampling fields per model family

  #401·xh123637 opened Nov 1, 2026·4 of 7 tasks·

  ·
  1

• feat(test-cases): add the bengali-web category (12 tasks) and a Bengali font for the base image

  #400·dipta007 opened Oct 31, 2026·5 of 8 tasks·

  ·

• feat(browser-runtime): add Browser Use as a managed runtime

  #395·gregpr07 opened Oct 21, 2026·4 of 8 tasks·

  ·

• feat(test-cases): add the shipping-quotes category (12 tasks)

  #393·nahatav opened Oct 17, 2026Contributor·6 of 9 tasks·

  ·

• feat(test-cases): add the privacy-rights category (12 tasks)

  #391·nahatav opened Oct 17, 2026Contributor·5 of 9 tasks·

  ·

• docs: add a real-run demo, concise bilingual READMEs, and a brand kit

  #386·reacher-z opened Oct 11, 2026Collaborator·8 of 11 tasks·

  ·
  2

• feat(adapters): AssistantBench suite import and its answer metric as a third stage

  #379·nahatav opened Sep 23, 2026Contributor·6 of 9 tasks·

  ·
  1

• fix(reproduce): check task count and compare only requested rubrics

  #376·plox-sumit opened Sep 17, 2026·

  ·
  1

• fix(rescore): require a matching cache fingerprint, not just match is not None

  #368·AmirF194 opened Sep 13, 2026Contributor·4 of 7 tasks·

  ·
  1

  1

• feat(adapters): WebVoyager task loader via the answer-submit interception path

  #356·vaibhavdabas16 opened Sep 12, 2026Contributor·

  ·

• feat(batch): --auto-restart supervisor and clawbench-batch-watch alerting

  #355·vaibhavdabas16 opened Sep 12, 2026Contributor·

  ·

• refactor(interceptor): one implementation of Stage-1 matching, with a parity matrix

  #354·vaibhavdabas16 opened Sep 12, 2026Contributor·

  ·
  1

• fix(harbor): align task schema, safe staging, and judge status

  #353·reacher-z opened Sep 10, 2026Collaborator·

  ·

• fix(harbor): make generated tasks safe on remote sandboxes (e2b)

  #346·vaibhavdabas16 opened Sep 9, 2026Contributor·

  ·

• feat(adapters): add the task-source adapter foundation and clawbench-sources CLI

  #345·vaibhavdabas16 opened Sep 9, 2026Contributor·

  ·

• feat(browser-runtime): implement the Steel browser runtime

  #344·vaibhavdabas16 opened Sep 9, 2026Contributor·

  ·
  1

  3

• feat(batch): report Stage 1 and Stage 2 together, never one alone

  bugenhancement
  #340·vaibhavdabas16 opened Sep 2, 2026Contributor·

  ·
  1

  2

• fix(batch): resume from run-meta.json, not from log-file existence

  #329·vaibhavdabas16 opened Aug 31, 2026Contributor·8 of 11 tasks·

  ·
  1

  1

• fix(judge): collapse judge_llm.py onto judge.py's transport

  #319·vaibhavdabas16 opened Aug 21, 2026Contributor·7 of 10 tasks·

  ·
  1