feat(work-log): native app and browser icons, honest failure and idle states

[rynfar]

Work logs showed a generic glyph for every tool, so a Chrome step, a native app
launch and a shell command all looked alike. A failed call replaced its icon with
an alert mark, losing the only cue for what had failed. And an Antigravity
subagent batch kept its running summary after its parent turn ended, implying
work was still happening when the batch could no longer report anything.

What changed

• Native app and browser icons. Work log rows show the app's own artwork or
  the page's favicon, resolved server-side and cached per environment.
• Failed calls keep their icon. A failed command still looks like a command,
  with a muted tint; icons that cannot take a tint get a trailing mark instead.
• Idle batches stop claiming a result. A batch whose parent turn ended shows
  the explanation rather than a stale running summary, and the conversation
  summary counts it without a completion checkmark.

Adopted from T3 Code

• 6cf0c6ea55d281f65c80502ec1871b0adf472025 (#9093) native app and browser icons
• c3b8825bf476cbce5e061c0f99570cf1f6723b89 (#9606) preserve tool icons on failed calls
• 00f8b7c28056188e3c5630160806a0afe51c9010 (#9616) idle subagent batches

Pylon adaptations

• Favicon host guard kept. Pylon refuses to build favicon URLs for private
  and reserved hosts; upstream's new page-derived lookup has no such filter. The
  guard now covers it, because a work log can be read from a remote client where
  http://localhost:5173 resolves against the viewer's network rather than the
  environment that browsed the page. One upstream test asserted that localhost
  returns a URL; it now asserts the refusal.
• Prime tool detail stays suppressed. The new presentation fields flow for
  every provider, but detail-like fields remain behind Pylon's primeAgentTool
  gate.
• Queued and waiting survive. Upstream's extracted spawn summary reports a
  live run with nothing running as "working". Pylon distinguishes N queued,
  queued, N waiting and working; the helper now carries that, since a
  spinner that claims work nobody is doing is the defect this repo cares about.

Converged on upstream

• One shimmer, not two. Pylon's live row hand-rolled an IntersectionObserver
  plus visibilitychange, focus and blur listeners to gate a per-label
  shine, because no shared gate existed when it was written. visibleAnimation.ts
  now provides exactly that for every animation in the app, so this adopts
  upstream's sweeping overlay, registers its container with the shared gate, and
  deletes 28 lines of bespoke wiring.
• One spawn summary. Upstream's deriveAgentSpawnSummary preserves Pylon's
  coordinator-authoritative live rule, so Pylon's inline duplicate is gone.
• One lifecycle extractor. Upstream defines a local copy of
  extractWorkLogToolLifecycleStatus; Pylon's shared version in
  client-runtime already encodes the same idle-batch rule, so the local copy
  was dropped rather than shadowing it.

Fixed along the way

docs/user/providers-antigravity.md shipped in #440 with 22 "T3 Code" mentions,
because that file was taken wholesale from upstream. All of them are the running
product referring to itself and now read "Pylon". The two mentions in
docs/user/install.md are left alone: they describe installing beside T3 Code as
a separate product, which is accurate.

Verification

• Tests: 415 across seven suites — subagent runtime, work-log presentation,
  favicon, mobile thread activity, spawn summary, and both timeline suites.
• Types: all packages clean.
• vp check: 0 errors across 3367 files.
• Unused exports and files: knip:check clean.

Model and harness: Claude Opus 5 via Claude Code.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[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	13.7 KiB	13.9 KiB	+158 B (+1.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.1 KiB	7.2 KiB	+106 B (+1.5%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.6 KiB	6.7 KiB	+52 B (+0.8%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	57.3 KiB	58.0 KiB	+700 B (+1.2%)	66.4 KiB	✅
Codex	Live turn messages	8	8	0 (0.0%)	21	✅
Claude	Total thread wire	13.8 KiB	14.1 KiB	+268 B (+1.9%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.1 KiB	7.2 KiB	+108 B (+1.5%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.7 KiB	6.9 KiB	+160 B (+2.3%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	58.2 KiB	59.7 KiB	+1.5 KiB (+2.6%)	66.4 KiB	✅
Claude	Live turn messages	10	10	0 (0.0%)	21	✅

Baseline: ab49d53 · PR result: 5a0c8f3 · 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: 115.6 KiB
• Claude decoded thread snapshot: 116.3 KiB

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