All pull requests

Pull requests

• fix(agents): stop react(submit=False) telling the model to call submit()

  #124·naimenz opened Sep 26, 2026·

  ·
  1

• feat(action_monitor)!: v6 prompt — evading a real site's block scores at least 4

  #123·Reih02 opened Sep 25, 2026Contributor·

  ·

• feat(human_agent): let several people into one sandbox, with operators

  #122·spruceb opened Sep 24, 2026·

  ·
  6

• feat(action_monitor)!: v8 prompt — rule (c) turns on what is hidden, not that something is

  #121·Reih02 opened Sep 22, 2026Contributor·

  ·

• Fix eval_awareness and broken_env prompts to name required result fields

  #120·tomseimandi opened Sep 21, 2026·

  ·

• feat(codex_centaur): centaur-mode Codex CLI reachable over SSH

  #116·MichalBravansky opened Sep 17, 2026·

  ·

• feat(action_monitor)!: v7 monitor prompt — score rather than decide, stepwise reasoning, per-score scale

  #115·Reih02 opened Sep 17, 2026Contributor·

  ·

• Add reward_hacking_v2 and general_misbehavior_v1 scanners

  #111·tim-hua-01 opened Sep 15, 2026·

  ·

• Add forum swarms with shared or private task sandboxes

  #110·spruceb opened Sep 11, 2026·

  ·
  2

• Restore original task instructions after native compaction

  #109·spruceb opened Sep 11, 2026·

  ·
  2

• feat(action_monitor): record what the sandbox declares about internet access and its hosts

  #108·Reih02 opened Sep 2, 2026Contributor·

  ·

• remove extraneous info event messages

  #106·MeganKW opened Sep 1, 2026Contributor·

  ·
  1

  

• feat: add llm reducer to scanners and use it by default

  #91·pipmc opened Aug 4, 2026Contributor·

  ·
  3

  

• chore(deps): bump pillow, aiohttp, and cryptography to patched versions

  #89·pipmc opened Aug 4, 2026Contributor·

  ·
  

• feat(scanners): recursive transcript-summary scanner

  #82·trentshaines opened Jun 18, 2026·

  ·
  7

  

• feat: add progress_scorer to Setting and progress-gated submit

  #72·rasmusfaber opened Apr 18, 2026Contributor·6 tasks·

  ·

• Improve human_baseline SSH setup

  #70·rasmusfaber opened Apr 14, 2026Contributor·5 tasks done·

  ·

• feat(common): add task and artifact utilities

  #69·rasmusfaber opened Apr 14, 2026Contributor·

  ·
  30

• feat(common): add Setting protocol to metr-inspect-common

  #67·rasmusfaber opened Apr 14, 2026Contributor·

  ·
  30

• Feat/react with reviewed submit

  #63·vncntt opened Apr 2, 2026Contributor·

  ·
  3

• Add instrumental convergence scanner

  #62·pipmc opened Mar 25, 2026Contributor·5 tasks done·

  ·