Issues

• [RFC] Store agent calls as ATIF trajectories and train from trajectories

  area: core
  Human-Agent-Society/reef#1054 · Filter by author BobbyZhouZijian opened on Oct 31, 2026

• [Task] AutoEnvScaling calls its task writer the author; Reef calls it the Designer

  area: harnessstatus: needs-triagetype: task
  Human-Agent-Society/reef#1050 · Filter by author Benjamin-eecs opened on Oct 30, 2026
  

• [Bug] An AutoEnvScaling outage spends rounds and parents

  area: harnessbugstatus: needs-triage
  Human-Agent-Society/reef#1049 · Filter by author Benjamin-eecs opened on Oct 30, 2026
  

• [RFC] Reef as five services that coordinate through one shared store

  area: serviceenhancementRFC
  Human-Agent-Society/reef#1048 · Filter by author BobbyZhouZijian opened on Oct 30, 2026

• [Feature] AutoEnvScaling writes the tasks of one generation one at a time

  area: harnessenhancementstatus: needs-triage
  Human-Agent-Society/reef#1046 · Filter by author Benjamin-eecs opened on Oct 30, 2026
  

• [Bug] AutoEnvScaling variants of TMax seeds get three times their parent's time

  area: trainingbugstatus: needs-triage
  Human-Agent-Society/reef#1043 · Filter by author Benjamin-eecs opened on Oct 30, 2026
  

• [RFC] Run task streams through Reef: rounds of served tasks, training between rounds, and unrecorded test tasks

  enhancementRFCstatus: needs-triage
  Human-Agent-Society/reef#1038 · Filter by author quao627 opened on Oct 30, 2026

• [RFC] Make Harbor tasks from a namespace's record items, split by arrival order

  enhancementRFCstatus: needs-triage
  Human-Agent-Society/reef#1037 · Filter by author quao627 opened on Oct 30, 2026

• [Bug] Concurrent rollouts of one task cancel each other's E2B template build

  area: harnessbugstatus: needs-triage
  Human-Agent-Society/reef#1035 · Filter by author Benjamin-eecs opened on Oct 30, 2026
  

• [RFC] Evaluation tasks as records: id, instruction, and answer

  area: harnessenhancementRFCstatus: needs-triage
  Human-Agent-Society/reef#1026 · Filter by author quao627 opened on Oct 29, 2026

• [RFC] Let a training instruction set its job's parameters, and train uploaded tasks

  area: trainingenhancementRFCstatus: needs-triage
  Human-Agent-Society/reef#1023 · Filter by author BobbyZhouZijian opened on Oct 29, 2026

• [RFC] Restore optional GPU sharing for Slime training and SGLang inference

  area: trainingenhancementRFCstatus: needs-triage
  Human-Agent-Society/reef#1021 · Filter by author Chonghe-Jiang opened on Oct 29, 2026

• [Question] Any empirical comparison between turn-level and whole-trajectory SFT?

  area: corequestionstatus: needs-triage
  Human-Agent-Society/reef#1019 · Filter by author yanan1116 opened on Oct 29, 2026

• [RFC] One sample store per scenario, a buffer of references, and one function from the buffer to the job

  area: trainingenhancementRFCstatus: needs-triage
  Human-Agent-Society/reef#1009 · Filter by author BobbyZhouZijian opened on Oct 29, 2026

• [RFC] Keep the Git LFS artifact repository on Hugging Face

  area: artifactsenhancementRFCstatus: ready
  Human-Agent-Society/reef#1002 · Filter by author BobbyZhouZijian opened on Oct 29, 2026

• [RFC] Serve artifacts beside the active one, each at its own URL

  area: versioningRFC
  Human-Agent-Society/reef#997 · Filter by author BobbyZhouZijian opened on Oct 29, 2026

• [RFC] Train full parameters on Spindle, and train from a full-weight import

  area: trainingarea: versioningenhancementRFCstatus: needs-triage
  Human-Agent-Society/reef#996 · Filter by author BobbyZhouZijian opened on Oct 29, 2026

• [Bug] An artifact import can change after it lands, and reads its source before it checks the component

  area: artifactsbug
  Human-Agent-Society/reef#992 · Filter by author Benjamin-eecs opened on Oct 29, 2026
  

• [Bug] Rollout continuation serves an image request without the image

  area: trainingbugstatus: needs-triage
  Human-Agent-Society/reef#987 · Filter by author augustuszzq opened on Oct 28, 2026

• [RFC] Let installed Harbor agents reach an evaluation's model from a remote sandbox

  area: trainingRFC
  Human-Agent-Society/reef#986 · Filter by author BobbyZhouZijian opened on Oct 28, 2026

• [RFC] Import an artifact from a directory or a URI

  area: versioningenhancementRFCstatus: needs-triage
  Human-Agent-Society/reef#980 · Filter by author BobbyZhouZijian opened on Oct 28, 2026

• [Feature] Seed AutoEnvScaling from parent Harbor tasks, and convert TMax-15K tasks

  area: harnessenhancement
  Human-Agent-Society/reef#977 · Filter by author Benjamin-eecs opened on Oct 28, 2026
  

• [Feature] Keep the caller's input_ids for image samples in the scoring pass

  area: trainingenhancement
  Human-Agent-Society/reef#968 · Filter by author BobbyZhouZijian opened on Oct 28, 2026
  1

• [Experiment] Live comparison and research follow-up of generated-task rollout (from #833)

  area: trainingtype: experiment
  Human-Agent-Society/reef#967 · Filter by author BobbyZhouZijian opened on Oct 28, 2026

• [Feature] Serve an artifact for its evaluation on the engine runtimes (RFC #945 step 4)

  area: trainingenhancement
  Human-Agent-Society/reef#966 · Filter by author BobbyZhouZijian opened on Oct 28, 2026
  2