All pull requests

Pull requests

• fix(action_monitor): hide tool calls proposed after the one under review

  #127·by Reih02 was merged Oct 2, 2026Contributor·

  ·
  5

• feat(action_monitor): size each monitor's conversation budget from its own context window

  #126·by Reih02 was merged Oct 2, 2026Contributor·

  ·
  1

• fix(action_monitor)!: tag transcript turns so agent text cannot forge them

  #125·by Reih02 was merged Oct 3, 2026Contributor·

  ·
  27

• feat(action_monitor): record prompt, score and threshold on each verdict

  #119·by Reih02 was merged Sep 18, 2026Contributor·

  ·

• feat(action_monitor)!: score every tool by default

  #118·by Reih02 was merged Sep 18, 2026Contributor·

  ·

• feat(monitor_validation): re-runnable validation suite for the action monitor

  #117·by Reih02 was merged Sep 19, 2026Contributor·

  ·
  10

• feat(action_monitor): open every escalation card with a human-review notice

  #114·by Reih02 was merged Sep 17, 2026Contributor·

  ·
  1

• feat(human): record shell commands as transcript tool events

  #113·by manishshettym was merged Sep 17, 2026Contributor·

  ·
  4

  

• action_monitor: add a v6 monitor prompt that allows for task specific instructions

  #112·by james-sullivan was merged Sep 17, 2026Contributor·

  ·
  7

• feat(action_monitor)!: score submit by default

  #107·by Reih02 was merged Sep 1, 2026Contributor·

  ·
  2

• Suspend token accounting on monitor calls

  #105·by MeganKW was merged Aug 31, 2026Contributor·

  ·
  3

• fix(action_monitor): keep monitor tokens off the sample's budget

  #104·by Reih02 was closed Sep 1, 2026Contributor·Draft·

  ·

• feat(action_monitor)!: v5 prompt — stop reading backgrounding as monitor evasion

  #103·by Reih02 was merged Aug 24, 2026Contributor·

  ·
  6

• perf(action_monitor): cache the transcript, not just the prefix

  #102·by Reih02 was merged Aug 21, 2026Contributor·

  ·
  9

• feat(action_monitor): callers supply the monitor chain

  #101·by Reih02 was merged Aug 21, 2026Contributor·6 tasks·

  ·
  19

• feat(action_monitor): configurable knobs, the monitor role, and a decider that discounts its wait

  #100·by Reih02 was merged Aug 19, 2026Contributor·

  ·
  3

• feat(monitor_cli): the metr-monitor operator CLI, and an operator guide

  #99·by Reih02 was closed Aug 19, 2026Contributor·Draft·

  ·
  3

• docs: add the MIT license

  #98·by pipmc was merged Aug 11, 2026Contributor·

  ·
  

• feat: add per-action LLM approval monitor

  #96·by Reih02 was merged Aug 11, 2026Contributor·

  ·
  40

• Add allow_early_submit to react_with_handoff_submit

  #95·by nikolajurkovic was merged Aug 12, 2026Contributor·

  ·
  22

• feat: add optional tool output truncation to scanners

  #93·by pipmc was merged Aug 12, 2026Contributor·

  ·
  3

  

• fix(devcontainer): pin containerd to 2.2.6 so docker-in-docker starts

  #90·by pipmc was merged Aug 11, 2026Contributor·

  ·
  

• feat(agents): add metr_agents/triage tool-call approver

  #88·by revmischa was closed Aug 11, 2026Contributor·

  ·
  13

• feat: add react_with_handoff_submit agent

  #87·by MeganKW was merged Aug 4, 2026Contributor·

  ·
  10

• feat(claude_uplift): human_agent with Claude Code ready at login

  #86·by MichalBravansky was closed Jul 29, 2026·

  ·
  2