Issues

• Proposal: 12-task public-procurement-search category

  TIGER-AI-Lab/ClawBench#382 · Filter by author yaiidad opened on Sep 30, 2026

• judge_context.gold_answer is read by the answer judge and documented, but task.schema.json rejects it

  TIGER-AI-Lab/ClawBench#380 · Filter by author nahatav opened on Sep 23, 2026

• Proposal: isolated environmental-observation-retrieval suite with 12 distinct tasks

  enhancement
  TIGER-AI-Lab/ClawBench#375 · Filter by author yangziao56 opened on Sep 14, 2026
  2

• publicity: audit existing outreach PRs and make featured-in claims traceable

  enhancement
  TIGER-AI-Lab/ClawBench#372 · Filter by author reacher-z opened on Sep 13, 2026

• publicity: prepare a reproducible next-release demo and bilingual announcement kit

  enhancement
  TIGER-AI-Lab/ClawBench#371 · Filter by author reacher-z opened on Sep 13, 2026

• publicity: turn the model leaderboard call-to-action into an evidence-based submission guide

  enhancement
  TIGER-AI-Lab/ClawBench#370 · Filter by author reacher-z opened on Sep 13, 2026

• Next release: data safety, scoring correctness, recovery and package validation checklist

  enhancement
  TIGER-AI-Lab/ClawBench#366 · Filter by author reacher-z opened on Sep 13, 2026
  3

• [P2] CI: runtime-server changes select zero downstream harness image builds

  bug
  TIGER-AI-Lab/ClawBench#365 · Filter by author reacher-z opened on Sep 13, 2026

• [P2] rescore: truncated judge cache is permanently skipped instead of retried

  bug
  TIGER-AI-Lab/ClawBench#364 · Filter by author reacher-z opened on Sep 13, 2026

• [P1] rescore: stale judge caches are silently attributed to a newly requested judge model

  bug
  TIGER-AI-Lab/ClawBench#363 · Filter by author reacher-z opened on Sep 13, 2026

• [P1] reproduce: validate sample coverage and selected rubrics before reporting PASS/FAIL

  bug
  TIGER-AI-Lab/ClawBench#362 · Filter by author reacher-z opened on Sep 13, 2026
  1

  

• Harbor upstream: agree live-site oracle evidence and parity acceptance protocol

  TIGER-AI-Lab/ClawBench#359 · Filter by author reacher-z opened on Sep 13, 2026

• Harbor: preserve verifier evidence and scoring when Kernel cleanup fails

  TIGER-AI-Lab/ClawBench#358 · Filter by author reacher-z opened on Sep 13, 2026

• Harbor: register browser MCP tools for the default local runtime

  TIGER-AI-Lab/ClawBench#357 · Filter by author reacher-z opened on Sep 13, 2026

• Priority issues to work on for contributors

  TIGER-AI-Lab/ClawBench#342 · Filter by author Perry2004 opened on Sep 8, 2026

• Harbor: strict harbor/terminal-bench format compliance + e2b sandbox support

  TIGER-AI-Lab/ClawBench#331 · Filter by author reacher-z opened on Aug 31, 2026
  3

  

• Proposal: isolated academia-registry-v0 suite for evidence-grounded registry research

  question
  TIGER-AI-Lab/ClawBench#317 · Filter by author reacher-z opened on Aug 20, 2026
  2

• feat(eval): support DeepSeek Harness and run a reproducible ClawBench benchmark

  enhancement
  Feature

  TIGER-AI-Lab/ClawBench#309 · Filter by author reacher-z opened on Aug 19, 2026
  1

  

• Stage-1 interceptor matching is duplicated between runtime-server and edgebench_judge, and the live copy has zero tests

  TIGER-AI-Lab/ClawBench#301 · Filter by author reacher-z opened on Aug 18, 2026
  1

• judge_llm.py has diverged from judge.py (no Gemini support, wrong responses endpoint, 800-token cap) — collapse the duplication

  TIGER-AI-Lab/ClawBench#300 · Filter by author reacher-z opened on Aug 18, 2026
  1

• batch --resume skips failed jobs (log-file heuristic) and overwrites batch-summary.json with 'skipped' rows

  TIGER-AI-Lab/ClawBench#297 · Filter by author reacher-z opened on Aug 18, 2026
  1

• rescore: public reproducibility CLI defaults to a maintainer's home paths and silently no-ops elsewhere

  TIGER-AI-Lab/ClawBench#296 · Filter by author reacher-z opened on Aug 18, 2026
  1

• Always report both Stage-1 (interception) and Stage-2 (judged); raw interception overcounts ~2×

  enhancement
  TIGER-AI-Lab/ClawBench#243 · Filter by author reacher-z opened on Jul 2, 2026

• feat: adapter — run AssistantBench (realistic time-consuming live-web tasks) under the ClawBench harness

  enhancement
  TIGER-AI-Lab/ClawBench#188 · Filter by author reacher-z opened on May 20, 2026
  1

• feat: adapter — run WebVoyager (live-website, screenshot+LLM-judge browser agent) under the ClawBench harness

  enhancement
  TIGER-AI-Lab/ClawBench#190 · Filter by author reacher-z opened on May 20, 2026