Issues

• ACP harness: no shutdown request is sent when the agent deadline cancels a prompt, so the agent's compaction edges are lost

  PrimeIntellect-ai/verifiers#3039 · Filter by author damienen opened on Nov 2, 2026

• Default bash harness: rejected compaction summaries come out with Branch.trainable=True

  PrimeIntellect-ai/verifiers#3038 · Filter by author damienen opened on Nov 2, 2026

• NeMo Gym taskset scores rollouts flagged mask_sample instead of dropping them

  PrimeIntellect-ai/verifiers#3015 · Filter by author earakely-scale opened on Oct 30, 2026

• GEPA accepts rewards from failed evaluations

  PrimeIntellect-ai/verifiers#2977 · Filter by author ItsM0rty opened on Oct 25, 2026

• write_episode can't save or reload traces with routed-experts, sampling-mask or top-logprob arrays

  PrimeIntellect-ai/verifiers#2956 · Filter by author salma2vec opened on Oct 22, 2026

• A commit rejected for bad training metadata leaves its nodes in the graph; a retry adds a second trainable branch

  PrimeIntellect-ai/verifiers#2954 · Filter by author salma2vec opened on Oct 22, 2026

• deepwiki and scratchpad examples: substring rewards pass without calling the tool they're meant to test

  PrimeIntellect-ai/verifiers#2948 · Filter by author jimmy-holdout opened on Oct 22, 2026

• proposer_solver: the proposer's contract answer is never verified, so false or ambiguous problems earn max learnability

  PrimeIntellect-ai/verifiers#2947 · Filter by author jimmy-holdout opened on Oct 22, 2026

• Toolset method named run shadows Toolset.run(): the tool server crashes at launch, but the rollout fails only after the 180 s port timeout, with the cause hidden in the trace

  PrimeIntellect-ai/verifiers#2937 · Filter by author 0reilly opened on Oct 21, 2026

• rlm harness: cell assigning a ShellJob (r = await rlm.shell.run(...)) returns an empty tool result

  PrimeIntellect-ai/verifiers#2924 · Filter by author rasdani opened on Oct 20, 2026

• vf-replay drops the saved taskset config when any --taskset.* flag is passed

  PrimeIntellect-ai/verifiers#2895 · Filter by author puiji opened on Oct 18, 2026

• Eval stuck during RL due to prime-sandbox/verifiers/harbor bug

  PrimeIntellect-ai/verifiers#2814 · Filter by author nxphi47 opened on Oct 8, 2026

• parse_pytest_outcomes turns a captured log line into a phantom test

  PrimeIntellect-ai/verifiers#2810 · Filter by author awesome-pro opened on Oct 8, 2026
  1

• vf-replay loses an entire run to a torn last line of traces.jsonl

  PrimeIntellect-ai/verifiers#2801 · Filter by author awesome-pro opened on Oct 7, 2026
  1

• rollout.close() keeps running a scorer after the other one fails

  PrimeIntellect-ai/verifiers#2800 · Filter by author awesome-pro opened on Oct 7, 2026

• ReferenceJudge: words in the judge's explanation override its leading yes/no verdict

  PrimeIntellect-ai/verifiers#2737 · Filter by author kv-248 opened on Oct 1, 2026
  1

• best-of-n: finalize() ranks a crashed (never-scored) sibling as reward 0, so it can be marked best over scored attempts and decide pass_at_n

  PrimeIntellect-ai/verifiers#2682 · Filter by author shaurya416 opened on Sep 25, 2026

• Chat streams: an upstream error event followed by [DONE] is recorded as a successful assistant turn

  PrimeIntellect-ai/verifiers#2659 · Filter by author Chaoqi31 opened on Sep 23, 2026

• Env-server pool never notices a dead worker; a worker that fails at startup hangs the eval

  PrimeIntellect-ai/verifiers#2571 · Filter by author willccbb opened on Sep 9, 2026

• Served runs rebuild every task as the taskset's declared Task type, silently dropping per-case subclasses

  PrimeIntellect-ai/verifiers#2570 · Filter by author willccbb opened on Sep 9, 2026

• v1: how should Harbor tasks that provision services via ENTRYPOINT bring up the agent-side environment?

  PrimeIntellect-ai/verifiers#2321 · Filter by author SergeyArc opened on Aug 11, 2026
  4

• v1: apply agentic-judge task weights when an isolated solver fails

  PrimeIntellect-ai/verifiers#2197 · Filter by author hallerite opened on Jul 31, 2026
  2

• v1: harden capture_patch handling of pre-existing untracked filenames

  PrimeIntellect-ai/verifiers#2196 · Filter by author hallerite opened on Jul 31, 2026
  2

• v1: complete artifact-transport regression coverage and threat-model documentation

  PrimeIntellect-ai/verifiers#2194 · Filter by author hallerite opened on Jul 31, 2026

• Support getting each turn in a multi-turn task as single training sample

  PrimeIntellect-ai/verifiers#170 · Filter by author LuanBrt opened on Jul 28, 2025
  2