Tracking issue for the DTE-v2 recreation. Replaces this fork's #28 (closed as superseded) and the old upstream series: PRs Zoo-Code-Org#1336 Zoo-Code-Org#1338 Zoo-Code-Org#1354 Zoo-Code-Org#1355 Zoo-Code-Org#1356 Zoo-Code-Org#1359 Zoo-Code-Org#1361 Zoo-Code-Org#1366 Zoo-Code-Org#1379 and issues Zoo-Code-Org#1328 Zoo-Code-Org#1329 Zoo-Code-Org#1330 Zoo-Code-Org#1331 Zoo-Code-Org#1332 Zoo-Code-Org#1376 Zoo-Code-Org#1377 Zoo-Code-Org#1378 - all closed as superseded on 2026-09.

• Content source of record: verified union tree feat/dte-trial-all @ 27a2e97df (branch retained; tag dte-legacy/union). Never per-PR heads (B1/M7/m8 hazard).
• Findings ledger: plans/dte-gap-review.md - 37 items (1 blocker / 8 major / 17 minor / 11 nit), all dispositioned in section 8 of the plan below.
• Left open deliberately: upstream Anthropic adaptive-thinking models: thinking content invisible (display omitted) and thinking_tokens telemetry missing Zoo-Code-Org/Zoo-Code#1326 + fix(api): surface adaptive thinking display and thinking_tokens for Anthropic models Zoo-Code-Org/Zoo-Code#1327 (F1 adaptive-thinking display fix). Its content is NOT part of the union content source (verified: absent from both main and the union tree); it is a standalone CI-green fix that must merge independently - the redo picks it up from base.

Unit status (maintained as PRs open/merge)

Risk flags (quick reference — review pass 2026-09-04)

• Near 400 (hunk-tune at push; reserved split point ready): U3 ~385 · U7 ~400 · U9 ~394 · U10 ~396 · U14 ~394 · U17 ~400 · U25 ~385 · U32a ~380.
• External conflict watch: U22 — ChatTextArea.tsx anchors overlap open upstream feat(ollama): add reasoning effort selectors and gate on Enable Thinking Zoo-Code-Org/Zoo-Code#1345; if it lands first, rebase U22 onto it (§10 watch item).
• Hunk-surgery heavy (mixed pre-existing files — double-check per §5 step 2): U3/U10/U13 (Task.ts), U13 (NewTaskTool.ts), U22 (ChatTextArea.tsx), U9 (NativeToolCallParser.ts / presentAssistantMessage.ts), U16 (ChatView.tsx), U16/U17 (ChatView.spec.tsx).
• Locale-heavy (18-locale weight inflates a+d — poor merge-rule candidates): U12 (~162), U24a/U24b (~207 each), U33b (~198).
• Timing: wave-6 e2e addenda (U25–U30, U35, U36) open only after base units merge — longest lead time.

DTE Final Ship Plan — "Ship It All" (DTE-v2, finalized 2026-09)

This is the single execution source of truth. It supersedes plans/dte-400line-recreation-plan.md (kept as historical inventory) and folds in every actionable finding from plans/dte-gap-review.md. Everything below is final: content source, unit ledger, trigger order, decisions, and the ship-all acceptance gate.

• Base of record: upstream/main (re-verify per branch; wave-1 base b2f63d366 unless main advanced).
• Content source of record (binding): the verified union tree origin/feat/dte-trial-all @ 27a2e97df — file content via git show 27a2e97df:<path>, hunks via git diff <base>..27a2e97df -- <path>.
• Rule that replaces the old per-branch source: never extract from a per-PR head. Per-PR heads are internally inconsistent — b17373bb9 (dte-3 tip) lacks the dte-5/dte-7-lineage fixes, 6eba686c1 (dte-5 tip) ships the strict-schema-breaking new_task.ts (B1), and 65ee7b5e6 (dte-7 tip) lacks the A5 orchestrator hardening (M7) and the dte-3 12-line effort-rank test (m8). Only the union is complete.

Union-tree verification (ran 2026-09, results pinned)

1. Decisions (all resolved — no open questions block execution)

2. Execution policy (in order, max 3 in flight)

1. Strict trigger order: open units only in the §4 sequence. A unit opens when its stack base branch exists (stacked) or is merged (e2e/main units), and a slot is free.
2. Max 3 open DTE-v2 PRs at any time (all tracks combined). Open the next in sequence when a slot frees (merge, or reviewer-ready after the CodeRabbit loop). No bursts.
3. Stack hygiene: after a merge, rebase the next stacked unit onto merged main (shallow stacks). e2e/main units (U25a, U25–U30, U36) open only after their base units are merged and rebase onto plain main (maintainer policy).
4. CodeRabbit rate: ≤4 included reviews/hour — stagger pushes; burst commits auto-pause review (resume with /resume).
5. Subagent/worker parallelism: at most 3 concurrent workers, triggered in the same sequence as the PRs.
6. Merge rule (optional, evaluated at push time): two or more adjacent units in the same wave may ship as one PR when (a) their measured combined a+d ≤ 350 (headroom for amendments + locale weight), (b) they share no hunk-surgery file with overlapping ranges, and (c) the combined mutation-diff scope stays fully killable in one Stryker run. Record the merge in the DTE-v2 final ship plan - 39-unit 400-line redo of the DTE series (supersedes #28) #35 ledger (both unit rows → same PR #). Never merge across wave/stack boundaries or to rescue an over-budget unit — split instead.

3. Line-counting rule (binding)

1. Measure the standalone diff: git diff --shortstat <stack-base> <branch> — the number GitHub shows against that base.
2. Counted = additions + deletions. Binary (PNG baselines) excluded from arithmetic, listed in PR body.
3. Hard budget: ≤400 counted lines per PR. Measure before opening (dry-run git diff --shortstat on the staged commit), not after CI. If the measure lands ≥380, restructure before pushing — move the next natural describe block / hunk group (the reserved split point) into the next unit — and re-measure. Never approve around the budget.
4. Mutation-gate corollary: every changed coverable line needs its killing test in the same PR (mutation-diff gates on changed code). The amendment ledger (§7) tests are no exception.

4. Final unit ledger (39 units; sequence column = trigger order)

Targets are estimates from the old branches ± amendments; measure every unit at push time and hunk-tune if over (⚠ marks units estimated ≥380).

Wave 1 — base (base: main)

Wave 2 — dte-2 content (stacked after U1)

Wave 3 — dte-3 content (stacked after U5) · Wave 4 — dte-5 content (parallel track, stacked after U5's head)

Wave 5 — dte-4 content (stacked after U11's head)

Wave 6 — e2e addenda (open only after base units merged; rebase onto main; content: union tree)

Wave 7 — F7 content (stacked after U18's head)

Totals: 39 units (38 full + 1 micro). Old-series own content ≈10,800 lines + amendments ≈900 lines ≈ 11,700 counted lines ÷ 400 ≈ 29 minimum; 39 units reflects natural feature boundaries + the required splits.

Risk flags (quick reference — review pass 2026-09-04)

• Near 400 (hunk-tune at push; reserved split point ready): U3 ~385 · U7 ~400 · U9 ~394 · U10 ~396 · U14 ~394 · U17 ~400 · U25 ~385 · U32a ~380.
• External conflict watch: U22 — ChatTextArea.tsx anchors overlap open upstream feat(ollama): add reasoning effort selectors and gate on Enable Thinking Zoo-Code-Org/Zoo-Code#1345; if it lands first, rebase U22 onto it (§10 watch item).
• Hunk-surgery heavy (mixed pre-existing files — double-check per §5 step 2): U3/U10/U13 (Task.ts), U13 (NewTaskTool.ts), U22 (ChatTextArea.tsx), U9 (NativeToolCallParser.ts / presentAssistantMessage.ts), U16 (ChatView.tsx), U16/U17 (ChatView.spec.tsx).
• Locale-heavy (18-locale weight inflates a+d — poor merge-rule candidates): U12 (~162), U24a/U24b (~207 each), U33b (~198).
• Timing: wave-6 e2e addenda (U25–U30, U35, U36) open only after base units merge — longest lead time.

5. Per-PR workflow (identical for every unit)

1. Worktree: git worktree add -b feat/dte-v2-<seq>-<slug> C:\work\Zoo-Code-fork\wt-dte-v2-<seq> <stack-base> + pnpm install --frozen-lockfile.
2. Extract from the union tree (§0): whole-file git show 27a2e97df:<path> > <file>; mixed files via hunk-level surgery (git diff <base>..27a2e97df -- <path> → hand-separated patches / edit to old content minus later-unit hunks). Apply the unit's §4 amendments on top. git status --short after every staging step. Hunk-surgery double-check (mixed files): git diff <stack-base> HEAD -- <paths> against the 27a2e97df hunk set for the file — every expected hunk present, no later-unit hunk leaked in (those belong to the stacked unit), no stray edits to pre-existing content. The §9 content-fidelity check is the final net; this catches per-unit mistakes early.
3. Local gates (all before push):
   • pnpm --dir src exec eslint --prune-suppressions --max-warnings=0 <each touched src file> — suppression counts never increase.
   • pnpm check-types (repo root); targeted vitest suites for every touched package (run from the package declaring Vitest).
   • node scripts/find-missing-translations.js if locale files changed (18 locales, src/i18n and webview-ui/src/i18n where keys are used).
   • themes:check only if visual baselines were regenerated (Docker artifact; never hand-edit playwright/themes/*).
4. Line budget gate: git diff --shortstat <stack-base> HEAD → a+d ≤ 400 (§3); split if over.
5. Commit & push (one logical commit, conventional message) → open draft PR against <stack-base> with the §6 body (stack note + budget statement mandatory; D3 note for dte-4 units; amendment list with finding IDs).
6. CI (§7) all green, incl. mutation-diff (expect ~30 min) and codecov/patch.
7. CodeRabbit loop until review complete with no actionable comments (push fixes or reply with reasoning).
8. Undraft → human maintainer approves → merge.
9. Bookkeeping: unit ledger row (PR #, head SHA, final measured a+d), fork tracking issue DTE-v2 final ship plan - 39-unit 400-line redo of the DTE series (supersedes #28) #35, plans/dynamic-thinking-effort-plan.md §10; mark the old PR(s) this unit replaces (already closed as superseded — keep the link in the close comment thread).

6. PR body template

Part of the DTE-v2 stack (≤400-line redo of <old PR #>); this PR: <one-line scope>.

**Stack**: base = `<stack base>` (or `main`). Standalone diff vs that base:
**<A> additions + <D> deletions = <A+D> lines (≤400 budget).** GitHub's number vs `main` will
look larger until lower stack PRs merge — review the standalone range.

**Related issue**: #<n>
**Amendments folded in** (from `plans/dte-gap-review.md`): <finding IDs + one line each, or "none">

**Out of scope**: <what this unit deliberately leaves for later units>

**Pre-submission checklist**:
- [ ] CI locally: check-types, targeted vitest, eslint --prune-suppressions (no count increase)
- [ ] i18n: all 18 locales updated + find-missing-translations clean (if keys added)
- [ ] Tests accompany all changed lines (mutation gate: no changed line waits for a later PR)
- [ ] Visual baselines (if UI): committed from Docker/CI artifact, listed below
- [ ] Line budget measured vs stack base (number above)

**Binary files in this PR**: <PNG baselines, or "none">

**Verification**: <commands + local results, incl. e2e mock run if e2e files touched>

7. CI targets (upstream/main b2f63d366; re-verify per push)

Undraft gate: all of the above green + CodeRabbit review complete with no actionable comments + final comment posted (findings → what changed; local verification; standalone a/d numbers).

8. Gap-amendment ledger (disposition of all 37 gap-review items)

In-PR (verified at merge of the listed unit): B1→U13/U14 · M1→U31 · M2→U32b · M4→U25a · M5→U36 · M6→U23 · M7→U13 (union source) · M8→the split itself · m1→U3+U21 · m2→U18 · m3→U13 · m4→U8+U20 · m5→U20 · m7→U6 · m8→U7/U8 (union source — free) · m9→U1 · m10→U11 · m11→U23 · m13→U16 · m14→U33 · m15→U32a · m16→U26 · m17→U29/U30 · n1/n2→U12 (+U24a quality pass) · n3→U11 · n4→U22 · n6→U13/U18 · n7→U13/U15/U18 · n8→U33b (optional) · n11→this document.

Documented keeps (no code change): m6 (U9 PR body) · m12/D3 (dte-4 PR bodies) · n9 (house pattern) · n10 (documented choice).

Post-merge follow-ups (file in tracking issue easonLiangWorldedtech/Zoo-Code#35; the list below is embedded in #35's body — file the concrete PRs as they open and check the rows off there):

1. M3 — o3-family wire override fix (openai.ts handleO3FamilyMessage → resolveEffectiveReasoningEffort; drop the static as "low"|"medium"|"high" cast) + tests. ~40–60 lines.
2. n5 — shouldUseReasoningEffort capability-absent branch: compare selected effort vs capability/default (PoE 400 risk).
3. n8 — Ollama xhigh/max→"high" clamp note in the settings hint (if not folded into U33b).
4. m10 family — repo-wide aria-label localization sweep (pre-existing ApiConfigSelector/ModeSelector/… patterns).
5. n9 — event-based settle for e2e teardown sleeps (only on flake).

9. Ship-all acceptance gate ("it's all shipped" checklist)

• All 39 units merged to upstream/main; ledger rows carry final PR #, head SHA, measured a+d (each ≤400).
• Content fidelity: for every DTE file, git diff <pre-DTE main>..main -- <file> equals git diff 3ea9f63e6..27a2e97df -- <file> except the listed amendments — i.e. per-file git diff main..27a2e97df -- <file> is empty, or its non-empty delta matches the §8 ledger entry. No union content silently dropped; no per-PR-head-only shape (B1/M7/m8 classes) shipped.
• Every CI lane green on every merged PR (mutation-diff, e2e incl. the new U36 suite, visual 0 drift, codecov/patch).
• All §8 follow-ups filed in tracking issue DTE-v2 final ship plan - 39-unit 400-line redo of the DTE series (supersedes #28) #35 with severity + rough size. (List embedded in DTE-v2 final ship plan - 39-unit 400-line redo of the DTE series (supersedes #28) #35 body on 2026-09-04; concrete follow-up PRs to be filed as they open and rows checked off there.)
• Old 9 PRs + 8 DTE issues + upstream trial PR feat(dte): dynamic thinking effort (series PRs 1-5 + e2e addenda + F7) Zoo-Code-Org/Zoo-Code#1379 + fork [ENHANCEMENT] Dynamic Thinking Effort: AI self-decides per-turn reasoning effort (experimental) — design + implementation plan #28 (tracking) + fork PR feat(dte): dynamic thinking effort (series PRs 1-5 + e2e addenda + F7) #29 (trial) closed as superseded (2026-09-04), each close comment links tracking issue DTE-v2 final ship plan - 39-unit 400-line redo of the DTE series (supersedes #28) #35; feat/dte-trial-all branch retained (tag dte-legacy/union 27a2e97df created). Upstream Anthropic adaptive-thinking models: thinking content invisible (display omitted) and thinking_tokens telemetry missing Zoo-Code-Org/Zoo-Code#1326/fix(api): surface adaptive thinking display and thinking_tokens for Anthropic models Zoo-Code-Org/Zoo-Code#1327 (F1 adaptive-thinking display fix) left open by design — content not in the union content source; merges independently.
• Tracking issue DTE-v2 final ship plan - 39-unit 400-line redo of the DTE series (supersedes #28) #35 + plans/dynamic-thinking-effort-plan.md §10 updated with the full new PR table. (DTE-v2 final ship plan - 39-unit 400-line redo of the DTE series (supersedes #28) #35 created 2026-09-04 with the full plan + 39-unit status table; PR # / a+d cells fill in as units land.)

10. Phase 0 (one-time prep)

git fetch upstream main
git tag dte-legacy/union 27a2e97df     # content source of record
# legacy tips (history only): dte-1 5db5cf4c8, dte-2 77ec064d3, dte-3 b17373bb9,
#   dte-4 bc1c1e4b8, dte-5 6eba686c1, dte-3-e2e 0b335c682, dte-5-e2e 471490b44, dte-7 65ee7b5e6

• Tag dte-legacy/union 27a2e97df created locally on 2026-09-04 (local reference; the branch feat/dte-trial-all on the fork remains the shared pointer).
• Done 2026-09-04: fork issue [ENHANCEMENT] Dynamic Thinking Effort: AI self-decides per-turn reasoning effort (experimental) — design + implementation plan #28 closed as superseded; new tracking issue easonLiangWorldedtech/Zoo-Code#35 opened with this plan + the gap-review ledger embedded in its body.
• Watch item: feat(ollama): add reasoning effort selectors and gate on Enable Thinking Zoo-Code-Org/Zoo-Code#1345 (chat-model-thinking-selector) is still OPEN and touches the U22 ChatTextArea.tsx anchors — if it merges first, rebase U22 onto it (policy recorded in old feat(webview): thinking effort surfaces — header chip, composer toggle, in-chat display (DTE series 4/5) Zoo-Code-Org/Zoo-Code#1359 body).
• Upstream main advance (2026-09-06): 0dbd5846f → 4140c2c83 — 2 infra commits: 4140c2c83 ([Fix] Commits fail after users interrupt mutation tests Zoo-Code-Org/Zoo-Code#1525) stryker interrupt fix (scripts/stryker-diff.mjs / .test.mjs / stryker.config.mjs tempDirName / mutation-testing.yml + .stryker-tmp eslint/tsconfig exclusions) and 057dfeebb ([Chore] Bump GitHub Actions to node24 runtimes to fix Node 20 deprecation warnings Zoo-Code-Org/Zoo-Code#1534) node24 CI runtimes. No sync needed: no src/ business-code changes; all 3 open PRs remain MERGEABLE with CI green on the current heads (1521 e89cceddd 19/19, 1522 069c34b9a 19/19, 1523 1087fcbd4 20/20); the only file overlap is U1's own DTE-1 discovery-flag change in scripts/stryker-diff.mjs, which is orthogonal to the [Fix] Commits fail after users interrupt mutation tests Zoo-Code-Org/Zoo-Code#1525 interrupt fix. Re-verify on the next main advance or at merge time.
• Upstream main advance (2026-09-06): 4140c2c83 → 9d43817fd — 1 commit: 9d43817fd ([Fix] Tasks stall when interrupted subtasks resume Zoo-Code-Org/Zoo-Code#1470) "[Fix] Tasks stall when interrupted subtasks resume" — native tool-call parser stream isolation (NativeToolCallParser per-task scope API; 24 files incl. src/core/task/Task.ts, scripts/stryker-diff.test.mjs, src/test-utils/native-tool-call-stream.ts, package.json, src/eslint-suppressions.json; merged by zoomote[bot] 2026-09-05 23:34:46Z). No sync needed: file overlap on exactly 2 files, both with disjoint hunks and orthogonal semantics — src/core/task/Task.ts ([Fix] Tasks stall when interrupted subtasks resume Zoo-Code-Org/Zoo-Code#1470 hunks in the streaming-parser-scope region L3022–L3600; U3's own hunks in imports/fields/updateApiConfiguration/setRuntimeThinkingEffort/dispose/API-metadata call sites — different regions, different subsystems) and scripts/stryker-diff.test.mjs ([Fix] Tasks stall when interrupted subtasks resume Zoo-Code-Org/Zoo-Code#1470 removes the dangling preferDirectTestFiles import + its describe block and fixes the parseVitestTestFiles test at L25/L181/L200+; U1's only hunk is the DTE-1 discovery-flag assertion at ~L129 in buildManifest). All 3 PRs re-verified MERGEABLE after the advance with heads unchanged (1521 e89cceddd, 1522 069c34b9a, 1523 1087fcbd4) and CI green on the current heads. Re-verify on the next main advance or at merge time.
• Upstream main advance (2026-09-06): 9d43817fd → 8d296deef — 1 commit: 8d296deef ([Fix] Task history disappears when user reopens a task Zoo-Code-Org/Zoo-Code#1319) "[Fix] Task history disappears when user reopens a task" — task-persistence resume/reopen fix (17 files: src/core/task-persistence/* incl. new mergeMessageSnapshots.ts + readFileWithMissingRetry.ts, src/core/task/Task.ts, src/core/task/apiConversationHistory.ts, src/core/webview/ClineProvider.ts, packages/types/src/message.ts, e2e + task-lifecycle docs + src/eslint-suppressions.json; merged by zoomote[bot] 2026-09-06 03:26:04Z). No sync needed: file overlap on exactly 1 file, with disjoint hunks and orthogonal semantics — src/core/task/Task.ts ([Fix] Task history disappears when user reopens a task Zoo-Code-Org/Zoo-Code#1319 hunks in the import block (L114), the resume/state region L982–L1240, L2111–L2227, and one line at L2676 (coords at 9d43817; identical to merge-base 0dbd5846f coords since [Fix] Tasks stall when interrupted subtasks resume Zoo-Code-Org/Zoo-Code#1470's Task.ts changes are all ≥L3022) vs U3's hunks at L24 (import), L320–L327 (fields), L1635–L1726 (updateApiConfiguration/setRuntimeThinkingEffort), L1759–L1845 (metadata), L2527–L2617, a pure insertion at L2619, and API call sites L4202/L4428/L4593). git merge-tree --write-tree upstream/main HEAD simulation is conflict-free for all 3 PRs (1521 → merged tree 573f3fc11, 1522 → bc279fd3a, 1523 → 666b4347b); heads unchanged (1521 e89cceddd, 1522 069c34b9a, 1523 1087fcbd4) and CI green on the current heads. Re-verify on the next main advance or at merge time.
• Bookkeeping (non-blocking): the verbatim reviewer 400-line quote still isn't pinned; locate it via gh api repos/Zoo-Code-Org/Zoo-Code/pulls/{N}/comments (old-series PR/issue threads are preserved after the 2026-09-04 closes — searchable via the close comments) and pin it in the base plan's §9; cross-reference in tracking issue DTE-v2 final ship plan - 39-unit 400-line redo of the DTE series (supersedes #28) #35. Counting stays additions+deletions regardless.