Skip to content

feat(tui): readable thinking state, redesigned Agents panel, and Agent Inspector#3108

Draft
aheritier wants to merge 3 commits into
mainfrom
feat/tui-agents-panel-thinking
Draft

feat(tui): readable thinking state, redesigned Agents panel, and Agent Inspector#3108
aheritier wants to merge 3 commits into
mainfrom
feat/tui-agents-panel-thinking

Conversation

@aheritier

@aheritier aheritier commented Jun 13, 2026

Copy link
Copy Markdown
Contributor

Summary

Overhauls the TUI sidebar Agents panel for readability and introduces an Agent Inspector. Three themed commits, one per area:

  1. Clearer thinking-state vocabularyCloses TUI: thinking-state indicator is ambiguous (reads as the model being on/off) #3105
  2. Redesigned Agents panel (focus card + compact roster)Closes TUI: redesign the sidebar Agents panel for large / parallel agent teams #3106
  3. Agent Inspector (right-click details modal)Closes TUI: add an Agent Inspector modal (configured + live agent details) #3107

What changed

  • Effort gauge (6-cell, lossless) replaces the ambiguous • on/off; adaptive→auto, token budgets show ◉ <count>; model names left-truncate to keep their tails.
  • Sidebar shows the current agent as a focus card in place and others as two-line compact rows with a width-aware degradation ladder and blank-line separators; harness agents show their harness type.
  • Right-click Agent Inspector (left-click switches, Ctrl+click also opens) showing full configuration + live per-toolset status; raw per-agent config is now retained on the team to source it.
  • Shared toolcommon helpers; formatTokenCount de-duplicated; Shift+Tab now shows a thinking-level toast.

Testing

  • task build, task lint clean; go test ./pkg/... green.
  • The only failing test is the pre-existing, network-only TestLoadExamples in pkg/teamloader (DMR model pull HTTP 416), unrelated to this change.
  • 9 new test files cover the gauge, panel/roster, click routing, inspector data, and config retention.

Demo

Screen.Recording.2026-06-13.at.22.56.14.mov

…elpers)

Replace the ambiguous bare on/off thinking label with a self-describing
visual language: a 6-cell lossless effort gauge, an adaptive->auto and
token-count label split, and left-truncation that preserves model-name
tails. Adds shared toolcommon helpers (effort_gauge, thinking, tokencount,
truncate) and de-duplicates formatTokenCount. Groundwork for the
agents-panel redesign and the agent inspector.
…ster)

Render the current agent as a focus card in place and the other agents as
two-line compact rows (name + effort gauge, then the full model id),
separated by blank lines, with a width-aware degradation ladder and
left-truncated model tails. Harness-backed agents show their harness type
instead of a blank model. Reworks the click-zone mapping to explicit
per-line ownership so clicks keep resolving to the right agent.
…ails

Add a read-only agent-details modal opened by right-click (or Ctrl+click);
plain left-click switches agent. The inspector shows the agent's
configuration (model, fallback, thinking, sub-agents, handoffs, skills,
limits, options, toolsets with their declared tools, commands) and live
state (current-agent marker and per-toolset started/stopped/error status
with live tool names when started). Retains the raw per-agent config on the
team and adds per-agent toolset-status and config accessors to source it.
@aheritier aheritier self-assigned this Jun 13, 2026
@aheritier aheritier requested review from dgageot, krissetto and rumpl June 13, 2026 20:59
@aheritier

aheritier commented Jun 13, 2026

Copy link
Copy Markdown
Contributor Author

I had some time to kill @rumpl @krissetto @dgageot
Not sure if there are some useful ideas

@aheritier aheritier added area/agent For work that has to do with the general agent loop/agentic features of the app area/tui For features/issues/fixes related to the TUI kind/feat PR adds a new feature (maps to feat: commit prefix) labels Jun 13, 2026
@aheritier

Copy link
Copy Markdown
Contributor Author

/review

@docker-agent

Copy link
Copy Markdown

PR Review Failed — The review agent encountered an error and could not complete the review. View logs.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

area/agent For work that has to do with the general agent loop/agentic features of the app area/tui For features/issues/fixes related to the TUI kind/feat PR adds a new feature (maps to feat: commit prefix)

Projects

None yet

2 participants