feat(scheduler): add Trae App heartbeat host

[cocolord]

Summary

• add a first-class trae_app host surface, scheduler runtime profile, agent type, and --trae_app CLI path
• reuse the provider-neutral App automation cadence, backoff, ACK, and liveness contracts while keeping Trae state isolated from Codex local automation storage
• preserve traex-cli and its existing visible /goal path as a separate host
• keep non-terminal settled heartbeat turns active for a fresh successor turn, and stop only on terminal outcomes
• preserve Codex App compatibility through a lossless legacy projection while making scheduler_hint.app_automation the canonical App protocol
• document the host boundary and add focused activation, parity, scheduler, thread-binding, and continuation coverage

Validation

• related Python tests: 715 passed
• TypeScript control-plane tests: 1235 passed, 1 skipped
• repository-configured python -m mypy: passed (22 source files)
• CI-scope Ruff (tests, loopx/canary, loopx/control_plane, loopx/domain_packs, loopx/presentation): passed
• git diff hygiene and DCO trailers: passed
• loopx check: public/private boundary clean for all candidate files

Pre-merge canary notes

loopx canary premerge --from-git-diff --git-diff-base origin/main selected 18 checks. Sixteen passed. The two remaining failures are non-feature gate issues:

• examples/control_plane/hot-path-interface-budget-smoke.py reports dashboard_status_json at 18,930 chars versus an 18,500 budget. Clean detached worktrees at both the earlier base and current origin/main produce the identical failure and size, confirming a baseline failure rather than a branch regression.
• examples/install-local-smoke.py reached the gate's fixed 120-second timeout. The exact-head GitHub stage2c (installed 0) job passed, so the install path is covered by the repository CI runner.

Exact-head GitHub CI completed successfully: DCO, dependency review, release/frontstage builds, dashboard acceptance, kernel static checks, Node minimum/forward compatibility, all four Python shards, installed-package, e2e, mutants, Windows, aggregate pytest, and Sonar passed. The Python Tests workflow concluded success; GitHub's check-run projection still displayed the already-completed merge-gate job as in_progress, although every job step, including the qualification rejection step, completed successfully.

Review notes

Changed surfaces: host activation/onboarding, scheduler execution context and hints, heartbeat prompt/liveness projection, quota ACK/failure state scoping, public protocol docs, and focused validation.

Only the real codex_app host may read the Codex local automation store. trae_app remains a distinct identity and state scope; it shares only the provider-neutral scheduler behavior. No benchmark jobs, scoring behavior, permission boundary changes, private state, raw logs, or generated evidence are included.

The bounded future-facing pass was applied at the existing typed App automation boundary; no second Trae-specific scheduler authority was introduced. Independent review is requested; this PR is not intended for self-merge.

[cocolord]

Request changes conclusion (author-owned PR; GitHub blocks formal self-review)

Reviewed exact head 5f4e9dbcf652014e946f7645a319be19b2ae3507 against origin/main@18fda32fda59e1ed1101c723eb3bfbc981b74bfd. Because this is an author-owned PR, this is posted as a COMMENTED review; the technical verdict is REQUEST_CHANGES and does not grant merge authority.

动机 / Motivation

这个 PR 解决的是一个真实且高频的连续执行缺口：Trae App 中一个 bounded turn 完成后，LoopX 没有 host-owned successor wake，剩余目标会停住，直到用户再次手动输入。补充独立的 trae_app host/runtime/state identity，并保留现有 traex-cli /goal 路径，是合理且必要的。

This PR addresses a real liveness gap: after a bounded Trae App turn settles, LoopX has no host-owned successor wake, so unfinished goal work can remain stranded until the user prompts again. Adding a distinct trae_app host/runtime/state identity while preserving the existing traex-cli /goal path is justified.

改动思路 / Approach

整体方向正确：复用现有 App cadence/backoff/ACK/liveness 算法，只在 host identity、thread binding、activation effect、scheduler-state surface 和 Codex-only local-store authority 上分流。这样避免实现第二套 scheduler，同时确保 Trae 不会读取或修改 Codex automation store。显式选择 trae_app 才会激活该路径，现有 Codex、generic CLI 和 traex-cli 默认行为保持隔离。

The overall design is sound: reuse the existing App cadence/backoff/ACK/liveness algorithm, while separating host identity, thread binding, activation effects, scheduler-state scope, and Codex-only local-store authority. This avoids a second scheduler implementation and keeps Trae away from the Codex automation store. The path is explicitly opt-in through trae_app; existing Codex, generic CLI, and traex-cli defaults remain isolated.

具体改动 / Changes

• SchedulerExecutionContext.app_automation_applicable and apply_scheduler_execution_context admit the new hosted App profile.
• _SchedulerHintBuilder.build emits Trae-scoped host facts and runnable ACK/failure commands without the Codex local fallback.
• _trae_app_activation creates the host-owned Trae automation activation packet.
• scheduler_ack._scheduler_packet reads the cadence packet and commits receipts under surface=trae_app.

The implementation threads the identity through guided start/onboarding, heartbeat prompt regeneration, quota decisions, state persistence, ACK/failure handling, ambient TRAECLI_THREAD_ID, documentation, and focused parity/isolation tests. The scope is broad but proportionate to the end-to-end host contract.

对主干的风险 / Main risk

[P1][blocking] trae_app 的公开 cadence packet 身份和 applicability 自相矛盾。

触发条件：以 runtime_profile=trae_app / --trae_app 执行 quota should-run，且存在需要 apply 或 ACK 的 App cadence transition。

路径：SchedulerExecutionContext.projection → apply_scheduler_execution_context → scheduler_hint.codex_app → scheduler_ack._scheduler_packet。首个矛盾可见于 loopx/control_plane/scheduler/execution_context.py:185。

同一个公开 packet 一方面声明 host_surface=trae_app 且 codex_app_applicability=not_applicable，另一方面又把 scheduler_hint.codex_app.applicability 标成 applicable，并继续使用 codex_app_* schema ids 和 scheduler_hint.codex_app.stateful_backoff。scheduler_ack._scheduler_packet 还显式把 trae_app 映射回 codex_app packet key。

结果是：按 execution context 行事的 consumer 可能因为 Codex 不适用而跳过 Trae 必须执行的 cadence update；按 packet key 行事的 consumer 又会把 Trae authority 误认为 Codex。当前测试 test_trae_app_runtime_profile_preserves_host_backoff_and_identity 先断言 Codex 不适用，再读取 hint["codex_app"]，因此把矛盾编码成了期望，而不是独立检出它。

Minimum repair: make the shared public cadence/applicability contract provider-neutral (for example, app_automation plus provider/host_surface) or emit a distinct trae_app packet. Update ACK/failure lookup, compact/effect consumers, and docs together; retain a bounded codex_app compatibility path only for actual Codex requests. Add a real quota-CLI regression that requires one self-consistent Trae packet identity and verifies that generated ACK/failure commands read and write only trae_app state.

Validation evidence:

• Real generated Trae ACK and failure commands both returned successfully, persisted under surface=trae_app, and failure replay produced host_update_failure_suppressed; no Codex local-store dependency was observed.
• The public-packet negative case failed as described above and is the blocking result.
• Exact-head CI run 34678346611 has 25/27 checks passing. The Windows failure and downstream merge-gate failure reproduce on origin/main run 34674379627, so they are baseline-equivalent rather than introduced by this diff.
• The focused suite passed 461/461 with the repository's venv interpreter available to subprocess tests. A later clean rerun reported 451 passed / 10 failed because the restricted child environment could not resolve that interpreter; all 10 are existing traex-cli subprocess cases, not changed-branch failures.
• No live Trae App automation was mutated during review. An external automation_update/readback integration remains residual validation after the protocol repair.

我的整体评价 / Overall assessment

REQUEST_CHANGES. 这项改动对 Trae App 明显是正向的：补齐了 successor heartbeat，边界选择合理，状态与 Codex local authority 也已隔离；代码量与问题规模相称。当前唯一确认的 blocker 是新公开协议把独立的 trae_app authority 装进 Codex-specific container，违反了本 PR 自己建立的一等身份。由于这条 API 尚未发布，现在修正成本最低。修复 packet identity、补齐真实 CLI contract test 并重跑风险对应验证后，我预期可以进入下一轮 approval review。

The change is directionally strong and proportionate: it closes the successor-heartbeat gap, preserves the existing scheduler owner, and isolates Trae state from Codex local authority. The single confirmed blocker is that the new public protocol puts first-class trae_app authority inside a Codex-specific container. This is the least expensive point to correct that API. After repairing the packet identity, adding the real CLI contract regression, and rerunning the risk-matched checks, I expect the PR to be ready for an approval pass.

[cocolord]

Approval conclusion (author-owned PR; GitHub blocks formal self-approval)

动机

Trae App 中的 LoopX goal 在完成一个有界阶段后，可能因缺少 host-owned successor wake 而停在非终态，表现为“只完成一部分、没有后续心跳推进”。主干此前只有 Codex App 的自动化路径和 traex-cli 的可见 /goal 路径，没有独立、可持久化且可回读的 trae_app host identity。

改动思路

把 trae_app 建模为独立的 typed host/runtime/state identity，同时复用现有 provider-neutral App automation 的 cadence、backoff、ACK、failure 和 liveness 规则。Canonical 协议统一为 scheduler_hint.app_automation；Codex 仅保留 lossless scheduler_hint.codex_app compatibility projection。Trae 不读取或写入 Codex 本地 automation store，也不接收 Codex alias/fallback。

具体改动

• 增加 HostSurface.TRAE_APP、SchedulerRuntimeProfile.TRAE_APP、--trae_app 和 TRAECLI_THREAD_ID 的完整入口与线程绑定。
• 将非终态 settled heartbeat 保持为 active，使 host 能创建下一轮；仅 terminal outcome 停止。
• 将 scheduler packet、state key、ACK/failure command 和 TypeScript persistence 按 surface 绑定；surface/state/operation 冲突时 fail closed。
• 保留 traex-cli 及其 /goal 路径，不把 Trae 冒充为 Codex。
• 保留 Codex v0 packet/reset-token compatibility，并用真实 CLI-to-TypeScript follow-up、compact envelope equality、fixed reset oracle 和 onboarding smoke 覆盖回归。

对主干的风险

主要风险是跨 host identity、scheduler state、compact envelope 与 compatibility projection 的联动回归。审查逐项核对了 typed state rules、domain neutrality、behavior-change disclosure 和 guidance-vs-obligation：没有 substring/prose-only 状态分类，没有把产品特定措辞放进通用控制面，也没有静默改变既有 host 的默认行为。trae_app 仅在显式选择时启用，Codex、SSH、generic CLI、goal runtime 与 traex-cli 维持原路径。

剩余边界是本次 review 没有实际修改用户的 live Trae automation；仓库内已覆盖生成命令、ACK/failure receipt 和持久化 readback。Premerge 的两个非 feature 失败已界定：install smoke 命中固定 120 秒超时，而 exact-head installed-package CI 通过；hot-path budget 的 18930 > 18500 在 exact base 上完全相同，不是本 PR 引入。

我的整体评价

对 exact base f88aae3a63c9a88fb2ddd3477f5b8dc3a05c2e0f 到 exact head 94b87611c1d41f9ac5a119a886e1ab46b11813b7 的 Standards 与 Spec 双轴审查均未发现可归因于本 PR 的 finding。结构化 review policy revision 3 一致性校验通过，结论为 APPROVE。由于作者不能在 GitHub 对自己的 PR 提交正式 approval，本 review 以 COMMENTED 发布；它不是合并动作。

验证证据：715 个相关 Python tests 通过；TypeScript control-plane 1235 passed、1 skipped；typecheck、repository-configured mypy（22 files）、CI-scope Ruff、diff hygiene、DCO、private/credential scans 和三个 focused smokes 通过。Exact-head GitHub workflow 已 completed/success，所有执行步骤成功；GitHub check-run projection 仍将已完成的 merge-gate 显示为 in_progress，属于远端展示状态漂移。

English verdict:

APPROVE for exact base f88aae3a63c9a88fb2ddd3477f5b8dc3a05c2e0f and exact head 94b87611c1d41f9ac5a119a886e1ab46b11813b7. The implementation gives Trae App a first-class trae_app host/runtime/state identity, reuses the provider-neutral App automation policy, preserves Codex compatibility without leaking Codex authority or storage into Trae, and keeps traex-cli separate.

Independent Standards and Spec reviews found no attributable issues. Typed state classification, domain-neutral control-plane wording, explicit behavior disclosure, and machine-enforced obligations were all checked. The relevant Python, TypeScript, type, lint, smoke, boundary, and exact-head CI evidence passed, subject only to the documented base-equivalent/local-timeout premerge exceptions. The GitHub workflow itself concluded successfully; the lingering merge-gate in-progress label is a stale check-run projection after all of its steps completed successfully.

This is an author-owned PR, so GitHub does not permit formal self-approval. I am therefore recording the approval conclusion as a COMMENTED review. No merge is performed by this review.

[huangruiteng]

动机

该 PR 为 Trae App 增加一等 heartbeat automation host，并把调度、ACK/failure、Turn receipt 和状态存储从 Codex 专属命名抽象为 provider-neutral 的 App automation，同时保持 Codex 兼容投影。目标是让 Trae 使用独立的 host/state identity，而不是误写 Codex 本地自动化存储。

改动思路

通过 HostSurface.TRAE_APP、SchedulerRuntimeProfile.TRAE_APP 和 APP_HEARTBEAT_SETTLEMENT_RUNTIME_PROFILES 建立类型化边界；scheduler_hint 生成 app_automation canonical packet，Codex 再由兼容层投影旧 codex_app 形状。CLI、heartbeat follow-up、onboarding 与 start-goal 复用同一 execution context，并在 host facts 中绑定 surface/state/reset identity。Trae 走 --trae_app 和独立 state key，Codex fallback 仅在 Codex host 上投影。

具体改动

关键代码讲解

• execution_context.py：新增 Trae host/profile、host automation applicability 和 guided-start/settlement 集合。
• scheduler_hint.py：统一构造 App cadence/backoff/ACK/failure；仅 Codex 通过 app_automation_compat.py 生成 legacy codex_app，Trae 不生成 Codex fallback。
• bootstrap_prompt.py 与 heartbeat_followup_cli.ts：支持 --trae_app，校验 host-facts 的 surface/state/RRULE 一致性，并拒绝 Trae 使用 Codex RRULE 别名。
• accountable_settlement.py：将 Codex/Trae 共同纳入 hosted-App settlement，保持 Turn receipt 与 quota settlement 绑定。

对主干的风险

发现两个合并前阻塞，均由真实 git merge-tree --write-tree origin/main origin/pr-4271-new 复现：loopx/control_plane/heartbeat/bootstrap_prompt.py 和 docs/reference/contracts/interface-budget-contract.md 均发生内容冲突。前者是 heartbeat prompt 的生成/解析入口，后者同时被 interface-budget smoke 和 host/capability 契约引用；直接选一侧会丢失另一侧的 bootstrap/Trae 参数或造成文档与 smoke 不一致。GitHub 当前也报告 mergeStateStatus=DIRTY，因此 exact head 不能安全合并。

验证方面，exact head 的 Trae/Codex scheduler Python focused subset 为 88 passed、128 deselected，六个变更相关 TypeScript suites 为 60 passed；更大 Python 集合在 342 个通过后因长时间等待手动中断。请先 rebase 当前 main，保留 canonical heartbeat/bootstrap 与接口预算契约，再重跑 host activation/onboarding、scheduler、interface-budget 及 packaged-surface smokes。

我的整体评价

Trae/Codex 隔离和 provider-neutral scheduler 方向正确，且 focused tests 支持主要路径；但冲突落在 prompt 生成器和接口预算这两个行为/验证入口，当前不能批准。请提交干净 rebase，确认 --trae_app prompt 可 round-trip、旧 Codex projection 不变，并完成上述风险对应验证后请求 re-review。当前结论：需要修改。

English verdict:

REQUEST_CHANGES: The Trae App boundary and Codex compatibility projection are coherent; 88 focused Python tests and 60 changed TypeScript tests pass. However, exact head 94b87611c1d41f9ac5a119a886e1ab46b11813b7 is DIRTY against current main, with real conflicts in loopx/control_plane/heartbeat/bootstrap_prompt.py and docs/reference/contracts/interface-budget-contract.md. Rebase, reconcile both canonical contracts, rerun host/bootstrap/interface-budget and packaged-surface smokes, then request re-review.

[huangruiteng]

动机

该 PR 为 Trae App 增加一等 heartbeat automation host，并把调度、ACK/failure、Turn receipt 和状态存储从 Codex 专属命名抽象为 provider-neutral 的 App automation，同时保持 Codex 兼容投影。目标是让 Trae 使用独立的 host/state identity，而不是误写 Codex 本地自动化存储。

改动思路

通过 HostSurface.TRAE_APP、SchedulerRuntimeProfile.TRAE_APP 和 APP_HEARTBEAT_SETTLEMENT_RUNTIME_PROFILES 建立类型化边界；scheduler_hint 生成 app_automation canonical packet，Codex 再由兼容层投影旧 codex_app 形状。CLI、heartbeat follow-up、onboarding 与 start-goal 复用同一 execution context，并在 host facts 中绑定 surface/state/reset identity。Trae 走 --trae_app 和独立 state key，Codex fallback 仅在 Codex host 上投影。

具体改动

关键代码讲解

• execution_context.py：新增 Trae host/profile、host automation applicability 和 guided-start/settlement 集合。
• scheduler_hint.py：统一构造 App cadence/backoff/ACK/failure；仅 Codex 通过 app_automation_compat.py 生成 legacy codex_app，Trae 不生成 Codex fallback。
• bootstrap_prompt.py 与 heartbeat_followup_cli.ts：支持 --trae_app，校验 host-facts 的 surface/state/RRULE 一致性，并拒绝 Trae 使用 Codex RRULE 别名。
• accountable_settlement.py：将 Codex/Trae 共同纳入 hosted-App settlement，保持 Turn receipt 与 quota settlement 绑定。

对主干的风险

发现两个合并前阻塞，均由真实 git merge-tree --write-tree origin/main origin/pr-4271-new 复现：loopx/control_plane/heartbeat/bootstrap_prompt.py 和 docs/reference/contracts/interface-budget-contract.md 均发生内容冲突。前者是 heartbeat prompt 的生成/解析入口，后者同时被 interface-budget smoke 和 host/capability 契约引用；直接选一侧会丢失另一侧的 bootstrap/Trae 参数或造成文档与 smoke 不一致。GitHub 当前也报告 mergeStateStatus=DIRTY，因此 exact head 不能安全合并。

验证方面，exact head 的 Trae/Codex scheduler Python focused subset 为 88 passed、128 deselected，六个变更相关 TypeScript suites 为 60 passed；更大 Python 集合在 342 个通过后因长时间等待手动中断。请先 rebase 当前 main，保留 canonical heartbeat/bootstrap 与接口预算契约，再重跑 host activation/onboarding、scheduler、interface-budget 及 packaged-surface smokes。

我的整体评价

Trae/Codex 隔离和 provider-neutral scheduler 方向正确，且 focused tests 支持主要路径；但冲突落在 prompt 生成器和接口预算这两个行为/验证入口，当前不能批准。请提交干净 rebase，确认 --trae_app prompt 可 round-trip、旧 Codex projection 不变，并完成上述风险对应验证后请求 re-review。当前结论：需要修改。

English verdict:

REQUEST_CHANGES: The Trae App boundary and Codex compatibility projection are coherent; 88 focused Python tests and 60 changed TypeScript tests pass. However, exact head 94b87611c1d41f9ac5a119a886e1ab46b11813b7 is DIRTY against current main, with real conflicts in loopx/control_plane/heartbeat/bootstrap_prompt.py and docs/reference/contracts/interface-budget-contract.md. Rebase, reconcile both canonical contracts, rerun host/bootstrap/interface-budget and packaged-surface smokes, then request re-review.

[huangruiteng]

动机

该 PR 为 Trae App 增加一等 heartbeat automation host，并把调度、ACK/failure、Turn receipt 和状态存储从 Codex 专属命名抽象为 provider-neutral 的 App automation，同时保持 Codex 兼容投影。目标是让 Trae 使用独立的 host/state identity，而不是误写 Codex 本地自动化存储。

改动思路

通过 HostSurface.TRAE_APP、SchedulerRuntimeProfile.TRAE_APP 和 APP_HEARTBEAT_SETTLEMENT_RUNTIME_PROFILES 建立类型化边界；scheduler_hint 生成 app_automation canonical packet，Codex 再由兼容层投影旧 codex_app 形状。CLI、heartbeat follow-up、onboarding 与 start-goal 复用同一 execution context，并在 host facts 中绑定 surface/state/reset identity。Trae 走 --trae_app 和独立 state key，Codex fallback 仅在 Codex host 上投影。

具体改动

关键代码讲解

• execution_context.py：新增 Trae host/profile、host automation applicability 和 guided-start/settlement 集合。
• scheduler_hint.py：统一构造 App cadence/backoff/ACK/failure；仅 Codex 通过 app_automation_compat.py 生成 legacy codex_app，Trae 不生成 Codex fallback。
• bootstrap_prompt.py 与 heartbeat_followup_cli.ts：支持 --trae_app，校验 host-facts 的 surface/state/RRULE 一致性，并拒绝 Trae 使用 Codex RRULE 别名。
• accountable_settlement.py：将 Codex/Trae 共同纳入 hosted-App settlement，保持 Turn receipt 与 quota settlement 绑定。

对主干的风险

发现两个合并前阻塞，均由真实 git merge-tree --write-tree origin/main origin/pr-4271-new 复现：loopx/control_plane/heartbeat/bootstrap_prompt.py 和 docs/reference/contracts/interface-budget-contract.md 均发生内容冲突。前者是 heartbeat prompt 的生成/解析入口，后者同时被 interface-budget smoke 和 host/capability 契约引用；直接选一侧会丢失另一侧的 bootstrap/Trae 参数或造成文档与 smoke 不一致。GitHub 当前也报告 mergeStateStatus=DIRTY，因此 exact head 不能安全合并。

验证方面，exact head 的 Trae/Codex scheduler Python focused subset 为 88 passed、128 deselected，六个变更相关 TypeScript suites 为 60 passed；更大 Python 集合在 342 个通过后因长时间等待手动中断。请先 rebase 当前 main，保留 canonical heartbeat/bootstrap 与接口预算契约，再重跑 host activation/onboarding、scheduler、interface-budget 及 packaged-surface smokes。

我的整体评价

Trae/Codex 隔离和 provider-neutral scheduler 方向正确，且 focused tests 支持主要路径；但冲突落在 prompt 生成器和接口预算这两个行为/验证入口，当前不能批准。请提交干净 rebase，确认 --trae_app prompt 可 round-trip、旧 Codex projection 不变，并完成上述风险对应验证后请求 re-review。当前结论：需要修改。

English verdict: REQUEST_CHANGES: The Trae App boundary and Codex compatibility projection are coherent; 88 focused Python tests and 60 changed TypeScript tests pass. However, exact head 94b87611c1d41f9ac5a119a886e1ab46b11813b7 is DIRTY against current main, with real conflicts in loopx/control_plane/heartbeat/bootstrap_prompt.py and docs/reference/contracts/interface-budget-contract.md. Rebase, reconcile both canonical contracts, rerun host/bootstrap/interface-budget and packaged-surface smokes, then request re-review.

[huangruiteng]

Rebased and repaired against current main at a371a30bf5071ac7a1bd43d8d0a050406bd44de5; exact head is 4344a1b886208c6e391b71564ede02becce8e752.

本轮处理：

• resolved loopx/control_plane/heartbeat/bootstrap_prompt.py by retaining current v2 canonical heartbeat bootstrap and adding explicit trae_app identity emission/parsing, including a real CLI round-trip regression;
• resolved docs/reference/contracts/interface-budget-contract.md by retaining current-main dashboard ceiling and aligning the quota row with this PR's canonical app_automation + Codex compatibility projection (14,000 chars / 350 nested keys).

Validation:

• current-head bootstrap regression: 12 passed;
• current-head changed TypeScript suites: 61 passed;
• current-head hot-path interface-budget smoke: passed (quota 13,838/14,000, dashboard 18,930/19,500);
• pre-latest-main host/scheduler focused Python suite: 372 passed;
• changed-file Ruff: passed;
• host/onboarding, heartbeat prompt/quota, quota plan, scheduler compaction/policy, status/quota parity, packaged-install and update smokes: passed;
• source-checkout premerge canary: diff hygiene, compile, risk-profile and public/private boundary checks passed. The only catalog failure was examples/install-local-smoke.py timing out at its fixed 120s limit; the same timeout reproduces standalone and is recorded as a baseline/environment hold. No live Trae automation was mutated.

冲突入口与 Trae/Codex identity 已完成收敛，旧 Codex projection 与 traex-cli 路径保持隔离；请求重新检查后可按 exact head 合并。

[huangruiteng]

Approved for exact head 4344a1b886208c6e391b71564ede02becce8e752 against current main a371a30bf5071ac7a1bd43d8d0a050406bd44de5.

The requested rebase conflicts in the canonical heartbeat bootstrap and interface-budget contract are resolved. The Trae App bootstrap now round-trips its explicit host identity while preserving the current v2 Codex compatibility path; the quota budget contract matches the provider-neutral packet and current main dashboard ceiling.

Evidence: all GitHub required checks passed, including Sign-off, dependency review, build, kernel static checks, Node minimum/forward compatibility, dashboard acceptance, Windows PowerShell, four test shards, stage2c e2e/installed/mutants/correctness, pytest, and merge-gate. Local focused host/scheduler Python (372 passed before the final main fast-forward; 12 passed on the final head), TypeScript (61 passed), interface-budget, onboarding/heartbeat/scheduler smokes, changed-file Ruff, and install/update package smokes passed. The standalone install-local smoke timeout is a known environment/baseline hold and is not a GitHub check failure.

No live Trae automation was mutated. Ready to merge.