fix(server): show OpenCode task progress in the composer

[rynfar]

Problem

OpenCode threads never showed task progress. The composer Tasks tab, the turn plan summary, and the sidebar working line were all permanently empty — OpenCode was the only provider in Pylon without plan parity:

Adapter	Plan surface
Codex	native turn/plan/updated ✅
Claude	TodoWrite input + Claude Tasks ✅
Cursor	ACP ext cursor/update_todos ✅
Grok	ACP PlanUpdated ✅
Prime (ACP + managed daemon)	✅
OpenCode	none

Two causes, both in OpenCodeAdapter.ts:

1. OpenCode emits a structured todo.updated event that was already arriving at the adapter and already passing the parent-session gate — it just fell off the end of the event switch, which has no default. No turn.plan.updated was ever emitted.
2. toToolLifecycleItemType matched todowrite on its write branch and classified it as file_change, so every todo update was filed into the edit tool group and counted toward the edit total. OpenCode threads over-reported file edits.

Found while reviewing upstream pingdotgg/t3code#5760, which diagnoses the same bug in T3. That issue is still open with no upstream fix, so this is Pylon's own.

Fix

Handle todo.updated and classify todo tools ahead of the write branch. Entirely within the adapter boundary — no contract, decider, projector, or migration changes, since turn.plan.updated already flows end to end for five other providers.

Notes on the choices:

• Uses the v2 SDK's todo.updated event, not the tool-call input the upstream issue suggests sniffing. The event is structured, already in the subscribed Event union, and survives OpenCode changing its tool internals.
• Cancelled todos are dropped, not reported completed. Pylon's plan contract has no cancelled state, and claiming finished work that was abandoned is worse than omitting it.
• Unknown statuses settle to pending — the SDK types Todo.status as a bare string.
• A fingerprint suppresses duplicates. OpenCode re-emits todo.updated on every mutation; without this each step transition would write a redundant plan activity. Same approach Grok and Prime already use.
• Only the parent session's todos count. A delegated child session would otherwise overwrite the thread's plan.

Verification

vp test run apps/server/src/provider/Layers/OpenCodeAdapter.test.ts — 84 passed, including 4 new.

Each new test was mutation-tested rather than trusted for going green:

• Disabling the todo.updated case → the 3 plan tests fail.
• Removing the todo branch from classification → the classification test fails.

vp run -F t3 typecheck — 0 errors. vp lint clean on both changed files, verified live with a debugger probe rather than inferred from a silent exit 0.

---

Claude Opus 5 in Pylon.

---

^{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	13.3 KiB	13.2 KiB	−121 B (−0.9%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	−2 B (−0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−119 B (−1.8%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.5 KiB	54.7 KiB	−822 B (−1.4%)	66.4 KiB	✅
Codex	Live turn messages	9	9	0 (0.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−7 B (−0.1%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	−8 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	+1 B (+0.0%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.4 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	10	10	0 (0.0%)	21	✅

Baseline: abe70a7 · PR result: 0e0afd0 · 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: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

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