Issues

• Judge moments: opening flag and first-turn fallback edge cases (#61 review round 4)

  toolboxmd/agent-observer#62 · Filter by author lukemaj opened on Oct 12, 2026

• Attribute per-PR cost inside large multi-link planner trees

  toolboxmd/agent-observer#58 · Filter by author lukemaj opened on Oct 12, 2026

• Detect moments where the user could not understand an agent's message

  toolboxmd/agent-observer#52 · Filter by author lukemaj opened on Oct 11, 2026

• Sanitize user-supplied price-source URLs and basis text in published summaries

  toolboxmd/agent-observer#30 · Filter by author lukemaj opened on Sep 27, 2026

• Needs-a-look inbox: ranked suspicious-work feed with dismiss verdicts

  toolboxmd/agent-observer#28 · Filter by author lukemaj opened on Sep 27, 2026

• Report task completion time, agent durations, failures and recovery delay

  toolboxmd/agent-observer#25 · Filter by author lukemaj opened on Sep 24, 2026
  8

• Diagnose context growth and cache behavior (from #3)

  toolboxmd/agent-observer#22 · Filter by author lukemaj opened on Sep 24, 2026

• Passive recording hooks for gaps the native records leave (from #2)

  toolboxmd/agent-observer#21 · Filter by author lukemaj opened on Sep 24, 2026

• Count Claude Code prompts typed mid-turn (queue-operation) exactly once

  toolboxmd/agent-observer#15 · Filter by author lukemaj opened on Sep 24, 2026

• Try Jev (System One decision model) as the judge for the improvement loop and as a detector engine (v2, grill first)

  toolboxmd/agent-observer#7 · Filter by author lukemaj opened on Sep 20, 2026

• Usage metering across hosts and machines, and an advisory capacity feed to Model Router

  toolboxmd/agent-observer#6 · Filter by author lukemaj opened on Sep 20, 2026
  2