fix(composer): clarify Prime Agent harness controls

[rynfar]

Problem

Prime Agent’s session controls were flattened into the composer footer: subagent depth, resource browsing, and a bare reload icon sat beside everyday controls. The reload action gave no visible indication of when it was useful, while an inactive persistent goal appeared only as Idle and the read-only popover did not teach users that goals are started through conversation.

How it was fixed

• Web and desktop now group subagent depth, resource browsing, and resource reload under one Harness menu.
• Reload is labeled Reload commands and resources and explains that it is only needed after changing session commands, skills, or prompts.
• Idle goal state is presented as No goal. The goal surface says it is managed in chat, explains that goals continue work across turns, and gives the prompt pattern “Start a persistent goal to …” instead of showing empty usage metrics.
• Mobile keeps its horizontally scrollable secondary toolbar, but uses the same goal language and gives the formerly icon-only reload action a visible label.
• Prime Agent user documentation now describes the conversational goal workflow and the new Harness grouping.

Surfaces

Web and desktop share the new Harness menu. Mobile retains its native toolbar structure with aligned labels and guidance. This is presentation-only: provider capabilities, server authority, wire contracts, remote behavior, and connection modes are unchanged. Pending Inputs remain outside the menu because they are live state requiring immediate visibility; context and goal remain status indicators beside the send action.

Verification

• 9 focused tests across client-runtime, web, and mobile
• targeted lint on all changed TypeScript/TSX files — clean
• typechecks for @t3tools/web, @t3tools/client-runtime, and @t3tools/mobile — pass (one pre-existing Effect suggestion in client-runtime/src/relay/discovery.ts)
• git diff --check — clean
• isolated real Prime Agent daemon pass: started a disposable no-tools thread, verified one Harness trigger, all depth choices, enabled idle reload, reload guidance, No goal, chat-managed goal guidance, and hidden empty metrics
• CI: Check, Test, Mobile Native Static Analysis, and Release Smoke pass

UI Changes

Before — depth and reload compete for footer space; goal reads “Idle”

After — one Harness menu owns session settings and explains reload

After — the inactive goal state teaches the conversational workflow

Checklist

• This PR is small and focused
• I explained what changed and why
• I included before/after screenshots for UI changes
• No animation or timing behavior changed, so no video is required

Model: GPT-5.6 Sol in Pylon (Prime Agent harness)

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	12.6 KiB	12.6 KiB	+8 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.3 KiB	6.3 KiB	−3 B (−0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.3 KiB	6.4 KiB	+11 B (+0.2%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	51.8 KiB	51.8 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	16	16	0 (0.0%)	21	✅
Claude	Total thread wire	12.6 KiB	12.6 KiB	−8 B (−0.1%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.3 KiB	6.3 KiB	−3 B (−0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.3 KiB	6.3 KiB	−5 B (−0.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	52.7 KiB	52.7 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	16	16	0 (0.0%)	21	✅

Baseline: 29bc8ed · PR result: ad64564 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 99.8 KiB
• Claude decoded thread snapshot: 100.6 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.