feat(desktop): goal chip shows paused state, elapsed, tokens, pause/resume

[me2seeks]

Summary

The desktop goal chip rendered a paused goal exactly like a running one — same pulsing accent dot, same counter — and the only control was clear, even though the host protocol fully supports goal.control pause | resume and resumeFromControl restarts a paused goal without spending a model turn.

• goal:pause / goal:resume are wired through preload (window.maka.goal.pause/resume) and the session-domain IPC bridge via the shared optimistic-revision controlGoalWithRetry path.
• Status-aware chip: active goals pulse, waiting goals remain live/accent but do not pulse, and paused goals use a still warning dot. Each state has its own accessible label.
• Elapsed and token context: elapsed wall-clock time and tokensSpent / tokenBudget render when available; paused elapsed time freezes at pausedAt.
• Pause/resume buttons appear next to the existing clear kill switch — pause for active/waiting goals and resume for paused goals — with matching overflow entries and localized failure toasts.
• Status-refusal handling: an operation_conflict at an unchanged revision is rethrown as the host's transition refusal instead of being retried as a race.

Runtime Host remains the lifecycle authority. The renderer reads goal.get, refetches on the canonical goal-change broadcast, and projects only the three live states (active | waiting | paused). Closes #3024.

Before / after

Before — a paused goal was still announced and rendered as running, with only iteration count and clear:

After — paused has a distinct still warning state, frozen elapsed time, token usage, resume, and clear:

Review process

Two independent read-only review passes (first-principles/Occam and simplify-audit) found no P0-P2 issues. The propagation chain remains single-authority end to end: control → accepted host commit/revision bump → canonical goal-change → renderer refetch → chip projection.

The remaining P3 is an existing extreme retry edge: the final optimistic-conflict attempt does not re-query to distinguish a last-window status refusal or removed goal. It requires three consecutive revision races and does not block this UI fix.

Test plan

• packages/ui: 170/170 tests pass, including active, waiting, and paused goal-chip rendering.
• Waiting-label reverse mutation fails the new regression as expected.
• Desktop preload/main/renderer/storybook TypeScript configs pass.
• Static Storybook build passes; visual smoke passes all 133 stories, including active/waiting/paused goal states.
• git diff --check passes.

AI use

• No generative tool was used for implementation.
• Generative tooling was used and the result was reviewed and verified by the author.

Tool(s) and scope: Maka authored the initial goal-chip implementation in a014abe7b; OpenAI Codex authored the waiting-state and goal-control race remediations in af7517963 and assisted with its tests, review remediation, and PR documentation. The author reviewed and verified both contributions.

Final squash trailers: Generated-by: Maka and Generated-by: Codex

[coderabbitai]

📝 Walkthrough

Problem solved

The desktop goal chip did not distinguish running, waiting, and paused goals. It also lacked pause/resume controls and did not show available elapsed-time or token-usage data.

This PR:

• Adds goal:pause and goal:resume IPC handlers.
• Reuses optimistic-revision retry logic for pause, resume, and clear actions.
• Rethrows status refusals when the revision is unchanged.
• Renders active, waiting, and paused goal states.
• Freezes paused elapsed time at pausedAt.
• Displays elapsed time and token usage when available.
• Adds pause, resume, and clear controls with localized failure toasts.
• Adds Storybook scenarios and regression tests.

Source of truth

The PR extends the existing host projection and canonical goal-change broadcast path. It does not create a parallel goal-state authority. The local timer only refreshes displayed elapsed time, and the renderer refetches after canonical broadcasts.

Scope and complexity

The solution is the smallest coherent path described by the diff. Shared retry logic avoids separate pause, resume, and clear implementations. Shared SessionContextGoal types replace the previous inline ChatView contract.

The added complexity is required for:

• Three distinct goal statuses.
• Revision-safe control operations.
• Pause-boundary timing.
• Optional token metadata.
• Localized controls and failure handling.
• Storybook and regression coverage.

No deletion is apparent without weakening behavior or coverage. The existing clear path was simplified by delegating to controlGoalWithRetry.

Complexity delta

• Authorities: No new state authority. Host projections remain authoritative.
• States: Adds explicit active, waiting, and paused states.
• Branches: Adds status-dependent rendering and control branches.
• Configuration: Adds localized pause/resume copy. No new configuration source is shown.
• Public surface: Adds preload goal.pause and goal.resume, the session-domain retry method, and expanded SessionContextGoal fields.
• Test maintenance: Adds control, status, timing, token, Storybook, and UI regression cases.

Total maintenance complexity increases in UI branches and public contracts, but the increase is justified by the required behavior and shared retry implementation.

Validation

The PR reports passing unit tests, TypeScript checks, Storybook build, visual smoke tests, and git diff --check. These results are not independently verified from direct check output here.

Review-relevant risks

The diff changes user-visible goal controls, status labels, timing, token display, and localized error messages. These UI changes require independent human review under repository policy.

The diff changes public preload and session-domain contracts. These contract changes require independent human review under repository policy.

No security, licensing, release, or governance effect was identified in the current diff.

The person performing the merge must review the final diff. A maintainer makes the final determination.

Walkthrough

The desktop goal flow now supports pause and resume through revision-aware IPC operations. The renderer exposes live goal metadata, localized controls, elapsed time, token usage, and status-specific active, waiting, and paused presentations.

Changes

Goal pause and resume controls

Layer / File(s)	Summary
Revision-aware goal control flow apps/desktop/src/main/runtime-host-client.ts, apps/desktop/src/main/runtime-host-session-domains-ipc-main.ts, apps/desktop/src/main/__tests__/runtime-host-client-operations.test.ts	Generalized retry handling supports pause, resume, and clear actions. IPC handlers validate session IDs and delegate to the retryable client method. Tests cover refreshed revisions and terminal status refusals.
Bridge and live goal wiring apps/desktop/src/preload/bridge-contract.d.ts, apps/desktop/src/preload/preload.ts, apps/desktop/src/renderer/use-session-goal.ts, apps/desktop/src/renderer/app-shell.tsx, apps/desktop/src/renderer/locales/shell-copy.ts	The preload bridge exposes pause and resume methods. Live goal state is narrowed to active, waiting, and paused statuses. The shell passes timing and token data and handles localized pause/resume failures.
Status-aware goal indicator packages/ui/src/session-context-layer.tsx, packages/ui/src/conversation-copy.ts, packages/ui/src/chat-view.tsx, packages/ui/src/icons.tsx	The goal indicator renders status-specific labels, elapsed time, token usage, and pause/resume/clear controls. Paused elapsed time is frozen, while other live states refresh periodically.
Goal state examples and coverage apps/desktop/stories/app-shell.stories.tsx, packages/ui/src/__tests__/session-context-layer-goal.test.tsx	Stories and tests cover active, waiting, and paused goals, including progress, timing, token display, and control affordances.

Estimated code review effort: 3 (Moderate) | ~25 minutes

Merge Risk: 🟡 Moderate · up to f2753

The new goal controls and status display can temporarily show an outdated pause/resume state, fail to keep paused elapsed time frozen in some cases, or show an erroneous failure toast after rapid repeated activation. The PR should receive follow-up on these bounded correctness issues before merge.

Sequence Diagram(s)

sequenceDiagram
  participant GoalIndicator
  participant MakaBridge
  participant IPCHandlers
  participant DesktopRuntimeHostClient
  GoalIndicator->>MakaBridge: pause(sessionId) or resume(sessionId)
  MakaBridge->>IPCHandlers: invoke goal:pause or goal:resume
  IPCHandlers->>DesktopRuntimeHostClient: controlGoalWithRetry(sessionId, action)
  DesktopRuntimeHostClient->>DesktopRuntimeHostClient: refresh revision and retry conflicts

Loading

Possibly related issues

• maka-agent/maka-agent#3023 — Both changes address revision-aware goal pause, resume, and clear control operations.

Possibly related PRs

• maka-agent/maka-agent#3026 — Both implement revision-aware goal controls with retry and conflict handling in different clients.

Suggested reviewers: astro-han, jackwener, likun666661

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	The implementation satisfies issue #3024 by adding pause/resume wiring, status-aware rendering, elapsed and token data, frozen paused timing, and controls.
Out of Scope Changes check	✅ Passed	The code, tests, stories, localization, and screenshots directly support the linked issue and stated goal-chip objectives.
Ai Use Disclosure	✅ Passed	The PR selects generative tooling, names OpenAI Codex (Maka) and its scope, and both introduced commits contain the consistent Generated-by: Maka trailer.
Title check	✅ Passed	The title clearly and concisely summarizes the main goal-chip changes: paused-state display, elapsed time, tokens, and pause/resume controls.
Description check	✅ Passed	The description covers the change, linked issue, verification results, screenshots, AI use, and review notes; it is mostly complete despite omitting the checklist section.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}

[qodo-code-review]

Code Review by Qodo

Sorry, something went wrong

We weren't able to complete the code review on our side. Please try again manually by commenting /agentic_review on this PR.

[Astro-Han]

I reviewed exact head a014abe7b85065d829b4549de6aa554e3d9b9c9f, including the full diff, current CI, Runtime Host goal lifecycle, IPC/preload boundary, and available automated feedback.

The primary architecture is correct. Runtime Host remains the sole goal lifecycle authority; Desktop reuses the existing goal.control contract and canonical projection rather than creating a second state machine. Consolidating clear/pause/resume through the same optimistic-control seam is also the right direction.

One P2 remains in the user-visible state contract.

SessionContextLayer only distinguishes paused; waiting still receives the running aria label, accent styling, and a pulsing dot. Runtime Host uses waiting for evaluator-directed backoff, when the goal is not currently consuming a model turn, and the issue explicitly requires active, waiting, and paused to remain distinguishable. The current chip therefore tells both sighted users and assistive technology that a waiting goal is running.

Please add an explicit waiting presentation branch and a focused rendering regression. The exact color can remain a design choice, but the label and motion must not reuse running semantics.

A non-blocking P3 remains in the final optimistic-conflict path: after repeated conflicts, a final state rejection can be collapsed into generic revision_conflict, hiding that another client already reached the requested state. A final query or idempotent-target check would improve that behavior.

No PR split is needed; the lifecycle authority and IPC structure should remain as implemented.

Disclosure: Codex performed the read-only source, lifecycle, accessibility, test, CI, and feedback analysis. The human contributor remains responsible for verifying the findings and deciding whether to merge.

中文

Goal authority 和 IPC 方向正确。唯一 P2 是 waiting 仍显示为 running 并持续 pulse，违反 active/waiting/paused 三态契约。补齐 waiting 文案、样式和测试后即可继续。

[Astro-Han]

Thanks — the three gaps from #3024 are real (paused looked identical to running, no pause/resume control, no elapsed/tokens), and the fix is the right shape: the chip renders purely from the host projection (single authority — use-session-goal reads goal.get, goal-change canonical broadcast triggers refetch), the 30s ticker is only a local rendering clock, and controlGoalWithRetry generalizes the existing clearGoal seam rather than adding a parallel path. The status-refusal detection is sound: only accepted transitions bump the revision (host commits always bump; illegal pause/resume/clear return undefined without commit), so "conflict + revision unchanged = refused" holds with no double-detection path, and the tests drive the helper with shifted responses. No overlap with #3025/#3026 (different packages, zero shared files, independent). Pause boundary timing is correct (running = Date.now()-setAt, paused frozen at pausedAt-setAt; resume includes wall-clock paused time, documented and consistent with the TUI). CI is green.

Conclusion: PASS — all P3, nothing blocking.

P3 (optional): in apps/desktop/src/main/runtime-host-client.ts:1168 the final attempt's break skips the status-refusal and null-goal detection — on a refusal in the last attempt it throws revisionConflict instead of the host reason (same as old behavior, not a regression; and a goal removed exactly in the last window turns a silent return into a spurious toast — needs 3 consecutive revision races, practically unreachable); the story at apps/desktop/stories/app-shell.stories.tsx:904 only adds setAt, not onPause/onResume, so the new buttons have no visual regression coverage (unit tests only); there's no goal e2e spec — pause/resume is "genuinely new behavior" and the repo standard would want an e2e alongside unit+storybook; formatCompactTokenCount prints "1000k" for ≥1,000,000 (cosmetic).

---

AI-assisted review disclosure: this review was produced with AI assistance (pi review subagent on ollama-cloud/deepseek-v4-flash). The subagent traced the full chain (control → commit/revision bump → canonical projection → goal-change → refetch) and the refusal detection logic from the PR head source; P3 items are static observations, not observed failures. Please weigh these findings with your own judgment.

中文摘要（AI 辅助审查）

结论：PASS，全 P3。问题真实（#3024 三缺口：paused 与 running 视觉无差别、无 pause/resume 控制、无 elapsed/tokens；host 协议本就支持 goal.control，projection 本就携带 setAt/pausedAt/tokensSpent/tokenBudget——纯 surface 缺口），方案正确：数据权威单一（chip 纯渲染 host projection，goal-change 广播触发 refetch，30s ticker 只是本地渲染时钟），controlGoalWithRetry 泛化既有 clearGoal 走同一 seam 无并行路径；状态拒绝检测 sound（仅被接受的转移 bump revision——commit 恒 bump、非法 pause/resume/clear 返回 undefined 不 commit，故 "conflict 且 revision 未变=拒绝" 推断成立无双误判）。与 #3025/#3026 零文件重叠无冲突。pause 边界时间计算正确（running=Date.now()-setAt、paused 冻结于 pausedAt-setAt；resume 后含暂停墙钟时间已文档化且与 TUI 语义一致）。CI 全绿。P3（可选）：runtime-host-client.ts:1168 末次尝试 break 跳过 status-refusal/null-goal 检测（末次遇 refusal 仍抛 revisionConflict、goal 恰在末次窗口被移除时从静默变多余报错 toast——需连续 3 次竞态才触达，实际不可达）；stories 只加 setAt 未加 onPause/onResume，新按钮无视觉回归覆盖；无 goal e2e spec（pause/resume 属 genuinely new behavior，按仓库标准应补 e2e）；formatCompactTokenCount 对 ≥1,000,000 输出 "1000k"（纯外观）。

[Astro-Han]

Before this can merge: this is a UI/UX change (goal chip now shows paused state, elapsed time, token count, and pause/resume controls), so per the project's review gate it needs before/after screenshots — e.g. the goal chip as it was (running vs paused indistinguishable) and as it is now (paused badge, elapsed, tokens, pause/resume button). The review itself is PASS (all P3s), but the merge is gated on the screenshots. Please add them to the PR description and re-request review.

AI-assisted review disclosure: this comment was written by a human maintainer following the review-gate rule; no AI contributed to this content.

[Astro-Han]

Blocking merge pending before/after screenshots of the UI/UX change (see comment above).

Review superseded by the main review comment; screenshot request communicated via comment.

[me2seeks]

@Astro-Han The waiting state now has a distinct non-pulsing presentation and accessible label in d8a0d2c, with focused regression coverage. The before/after screenshots are also in the updated PR description and both image links are live. Could you please re-review the current head when convenient?

[Astro-Han]

Thanks for making the goal chip reflect the lifecycle users actually need to reason about. I reviewed the current head with an independent @reviewer pass and a read-only ollama-cloud/deepseek-v4-flash:high pass.

The problem is correctly defined: a goal is not merely active or finished; paused and waiting states must remain distinguishable, and pause/resume must go through the Runtime Host authority rather than renderer-owned state. The implementation follows that seam, carries the operations through the existing Host/IPC/preload boundary, and keeps the UI projection focused on elapsed time, token usage, and the permitted action.

I found no concrete P0–P2 issue on this head. The current CI checks are green. The remaining observations are non-blocking: the final conflict query/direct IPC path and a narrower end-to-end assertion could be pinned more explicitly, but they do not establish a production defect.

No local test suite was run during this review; conclusions are based on source, test, and current CI inspection. Codex coordinated the independent passes and performed the final adjudication; external-model output was treated as unverified until checked against the code.

中文摘要

感谢这次让 goal chip 真正反映用户需要理解的生命周期。问题定义正确：goal 不只是活跃或结束，paused 与 waiting 必须可区分，pause/resume 也应经过 Runtime Host 权威，而不是由 renderer 自建状态。

当前实现沿用了现有 Host/IPC/preload 边界，UI 只投影 elapsed time、token usage 和允许的操作。当前 head 未发现具体的 P0–P2 问题，CI 也已全绿。

剩余仅是非阻断测试建议：可以更明确地固定最终冲突查询/直接 IPC 路径，以及增加更窄的端到端断言，但目前没有可复现的生产缺陷。

本次未在本地运行测试套件；结论来自源码、测试与当前 CI 检查。Codex 协调了独立审查并完成最终判断，外部模型输出在核对代码前均视为未验证输入。

[Astro-Han]

Non-blocking design suggestion

The Runtime Host, IPC, and state-management implementation looks solid, so I do not consider this a merge blocker.

From a product UI hierarchy perspective, however, an active Goal feels less like session metadata and more like a live execution control: it can continue consuming a budget and may need to be paused, resumed, or cleared while the user is working in the composer.

I suggest a follow-up that moves the Goal presentation out of SessionContextLayer and into Astryx ChatComposer.headerContext. Keeping the status and kill switch next to the primary input should make the relationship clearer and produce a calmer, more cohesive layout.

A compact composition could keep StatusDot, progress/budget text, and pause/resume visible, while placing Clear in MoreMenu. I would avoid ChatComposerDrawer, because it represents collapsible staged input context, and Banner, because it would give an ordinary running state too much visual weight.

This is a design refinement rather than a correctness concern; the current implementation is functionally sound and can land as-is.

Disclosure: Codex assisted by inspecting the PR structure and the available Astryx component seams. Astro-Han independently reviewed the code and agrees with this non-blocking design recommendation.

中文

后端、Runtime Host、IPC 和状态管理实现没有问题，这条建议不阻塞合并。

从产品界面的信息层级看，Goal 更像持续运行的执行控制器，而不是会话元数据。建议后续将它从 SessionContextLayer 移到 Astryx ChatComposer.headerContext，让状态、预算和暂停入口靠近用户的主要操作区域。

可以始终展示 StatusDot、进度/预算以及暂停或恢复按钮，将 Clear 收进 MoreMenu。不建议使用可折叠的 ChatComposerDrawer，也不建议使用视觉权重过高的 Banner。

[Astro-Han]

One provenance clarification remains before merge. The PR body says OpenAI Codex (Maka) assisted implementation and review remediation, and commit a014abe7b carries Generated-by: Maka, but the follow-up substantive fix commit d8a0d2c90 does not.

If that waiting-state remediation used substantive AI assistance, please amend it with the matching Generated-by: Maka trailer and preserve the trailer in the final squash commit. If it was entirely human-authored, please clarify that exception in the PR body instead. The technical approval and accepted before/after screenshots otherwise stand.

中文说明

PR body 说明 Maka 参与了实现和 review remediation，首个 commit 也有对应 trailer，但后续 waiting-state 修复 commit d8a0d2c90 没有。

如果该修复使用了实质 AI 辅助，请补 Generated-by: Maka；如果完全由人工完成，请在 PR body 明确这个例外。技术审查和 before/after 截图结论不变。

[me2seeks]

Correction to my previous provenance reply: the two commits used different tools. a014abe7b was authored with Maka and already carries Generated-by: Maka; the waiting-state remediation was authored with OpenAI Codex, so I corrected its trailer to Generated-by: Codex in metadata-only head 91b50fc501829378d27d18ab1960e54d95628550. The tree remains exactly eba908780f668d1188d48eb525771285742039e5. The PR body now names each tool and commit separately, and the final squash should preserve both trailers.

[coderabbitai]

Actionable comments posted: 3

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 374eadb1-5335-46c9-b664-185951aa43a8

📥 Commits

Reviewing files that changed from the base of the PR and between 7bb776b and f2753e6.

📒 Files selected for processing (14)

• apps/desktop/src/main/__tests__/runtime-host-client-operations.test.ts
• apps/desktop/src/main/runtime-host-client.ts
• apps/desktop/src/main/runtime-host-session-domains-ipc-main.ts
• apps/desktop/src/preload/bridge-contract.d.ts
• apps/desktop/src/preload/preload.ts
• apps/desktop/src/renderer/app-shell.tsx
• apps/desktop/src/renderer/locales/shell-copy.ts
• apps/desktop/src/renderer/use-session-goal.ts
• apps/desktop/stories/app-shell.stories.tsx
• packages/ui/src/__tests__/session-context-layer-goal.test.tsx
• packages/ui/src/chat-view.tsx
• packages/ui/src/conversation-copy.ts
• packages/ui/src/icons.tsx
• packages/ui/src/session-context-layer.tsx

Included review availability: Your plan provides up to 3 included reviews per hour; 2 remain after this review.

[me2seeks]

The failed Runtime Host lane was an unrelated daily-review-two-client-uds timestamp race (generatedAt differed by 4 ms); the aggregate test only followed that lane. Contributor permissions cannot rerun repository jobs directly, so I metadata-only amended the existing Codex-authored commit to 731c8fedd and force-pushed it. The tree is unchanged (eba908780f668d1188d48eb525771285742039e5), and fresh CI is now running.

[me2seeks]

Latest bot-review remediation is on af7517963: goal reads are latest-request-wins, pause/resume controls are serialized per session, and paused UI state requires the Host-projected pausedAt timestamp. Local: UI full 170/170, Desktop typecheck, Biome, and diff checks pass. All three threads are replied and resolved.

[Astro-Han]

Reviewed exact head af751796328c27aeca0613c704e25223a8123e41.

The latest delta correctly projects the waiting state, sequences goal reads so stale responses cannot overwrite newer state, and serializes pause/resume per session. Keeping clear outside that guard is appropriate for the existing kill-switch semantics: it remains revision-safe and available while another control is pending. All checks are green, all review threads are resolved, the before/after screenshots demonstrate the paused-goal controls, and the Maka/Codex disclosure plus all three commit trailers are complete. I found no new P0–P3 issue.

Please preserve both declared provenance trailers in the final squash commit.

AI-assisted review disclosure: Codex reviewed the exact-head diff, control/projection concurrency, current-main integration, tests, live CI and threads, screenshots, and provenance metadata. No external model was used. Astro-Han authorized this review campaign.

中文说明

最新提交正确补齐了 waiting 状态、避免旧 goal 查询覆盖新状态，并按 session 串行化 pause/resume。clear 保持独立符合既有 kill switch 语义：它仍有 revision 安全控制，也能在其他操作 pending 时立即清除。CI、线程、before/after 截图以及三个提交的 Maka/Codex trailer 均已核实，没有新的 P0–P3。最终 squash 请保留两种 provenance trailer。