feat(tui): render durable task mutation history

[me2seeks]

Summary

• hydrate task_create and task_update cards from the Runtime Host durable Task mutation projection
• render up to 20 created tasks inline; summarize larger creates in the live TUI and show every item in /transcript
• keep pagination, history restarts, exact Session/revision validation, stale-request fencing, and immutable-scrollback notices inside the CLI adapter/controller
• batch correlations by both the 128-item limit and the Host canonical 192 KiB encoded-byte budget
• return hydration results only after every batch succeeds, so a later failure cannot publish partial Task history

Authority and scope

The TUI does not parse tool-result prose or reconstruct Task history from the current Task snapshot. It consumes the Host-owned immutable presentation keyed by (turnId, toolCallId) and fails closed on Session, revision, correlation, or operation mismatches. task_list and task_get remain unchanged.

This PR depends on #4239. Until #4239 merges, the GitHub diff also contains its Runtime Host protocol commits. The TUI commits are d8d0efa, 4d52b12, and 53593f1 on the current rebased head.

UX

• 1-20 created tasks: show the complete list at the original tool-call position
• 21+ created tasks: Added N tasks · /transcript to view full list
• updates: show the Task key, prior and next status, subject, and durable reason/evidence
• entries already frozen in terminal scrollback keep their rendered lines and receive one tail notice pointing to /transcript

Verification

• maka-agent full test suite on latest main: 664 passed, 0 failed
• focused Task mutation driver tests: 7 passed, including byte-aware start/continue batches and later-batch failure
• Runtime Host and CLI builds passed
• lint and format checks passed
• two independent pre-commit and exact-head reviews: 0 blockers

Part of #4179.

[M4n5ter]

I found three P1 blockers, one P2, and one P3 on exact head e0e5e4246c45082877c3e36af3e6d8da36dbf179. Two P1s are inherited from dependency PR #4239: the current-main compatibility-epoch collision and rejection of supported nested Code Mode tool-call identifiers. I documented those in the dependency review. The incremental P1 allows durable model-controlled task text to emit terminal control sequences. The incremental P2 repeatedly re-queries immutable history, and the P3 announces that unavailable details are ready. Details and fixes are inline.

I ran 95 focused tests across the TUI hydration/rendering suites, the complete Runtime Host Session driver suite, and the Task mutation protocol/projection suites; all passed. Production-seam probes separately reproduced the nested-ID rejection, preserved OSC-52 bytes in final terminal output, and the contradictory frozen notice. The exact-head test, windows_recovery, and label checks are green. The stack still conflicts with current main at the compatibility epoch, so merge-result validation must wait for the rebase.

---

Automated review notice: This comment was posted by an automated review agent operated by M4n5ter. It is not an independent human review and does not replace one.

[Astro-Han]

Thanks for rebasing and continuing to harden the Task mutation presentation.

Before we spend more review time on the current 3,700-line direction, I think we need to reconcile it with #2290. That RFC has already converged on SessionTodo as a narrow current snapshot and tracks eventual removal of the model-facing task_* tools, mutation-event history, paginated query protocol, claim/settle paths, and the current Task panel.

This PR moves in the opposite direction by adding a new durable Task mutation-history query, continuation contract, global validation, wire batching, and TUI hydration around the authority #2290 intends to retire.

Could we pause this PR until the SessionTodo contract is settled? If a semantic mutation timeline is still needed afterward, it can be designed against the smaller surviving authority rather than extending the current Ledger first. Please push back if #2290’s accepted direction has changed or this history has an independent obligation that survives the migration.

AI-assisted review disclosure: Codex assisted with comparing this exact head against #2290; Astro-Han verified the architectural conflict and owns this position.