[swingtempo]

Summary

• AutoComboCatalog - each autocombo card (e.g. auto/best-coding) has a button to generate a snapshot of the combo so a user can see and edit it
• Upon clicking, the combo gets generated
• window scrolls to the newly created card

Related Issues

• Closes [Feature] Allow duplication of auto-combos, materialize the list into a snapshot for easy editing #10231 (looks like Feature team already closed it and tracking is as an internal feature reqeusts)
• Video in discussion: Snapshot an auto combo so user can tweak it #10355

What Changed

1. Snapshot Weight Computation (open-sse/services/autoCombo/virtualFactory.ts)

• New function computeSnapshotWeights() scores candidates at combo creation time using:
  • taskFit: reasoning & vision-capable models score higher when taskFit is weighted
  • stability: richer capability profiles assumed more stable
  • tierPriority / costInv: premium tiers and free models get bonuses based on the weight pack
  • latencyInv / health / quota: neutral baseline (no runtime telemetry at snapshot time)
• Replaces hardcoded weight: 1 in createVirtualAutoComboFromPrepared() with computed scores normalized to [0, 1].

2. Duplicate Endpoint (src/app/api/combos/duplicate/route.ts) — NEW

• POST /api/combos/duplicate resolves an auto/* template (e.g. auto/best-coding, auto/glm) into a static combo snapshot.
• Reuses the same candidate resolution logic as createVirtualAutoCombo().
• Normalizes weights to sum exactly 100, applies naming convention (static-{name} copy), and preserves the original weight pack in config.
• Returns 201 on success with the created combo object; validates input (400) and unknown templates (422).

3. Dashboard UI Updates

File	Change
src/app/(dashboard)/dashboard/combos/page.tsx	Added handleComboCreated() callback to auto-scroll newly created combo cards; wired it into <AutoComboCatalog />. Refactored ComboFormModal provider builder grid layout.
src/app/(dashboard)/dashboard/combos/AutoComboCatalog.tsx	Accepts optional onComboCreated prop for post-creation scroll behavior.

4. Tests (Rule #18)

Test File	Coverage
tests/unit/snapshot-weights.test.ts	Capability differentiation, tier bonuses, neutral baselines, score clamping
tests/unit/combos-duplicate-route.test.ts	Auth gate, input validation (400), success response shape, weight distribution, naming convention, error sanitization
tests/unit/combos-duplicate-resolution-audit.test.ts	Audit-style checks for duplicate resolution edge cases

5. Utilities & i18n

• scripts/ad-hoc/dump-auto-combos.ts: CLI helper to dump resolved auto-combo candidates (debug / inspection).
• ~47 locale files: minor string updates in src/i18n/messages/*.json.

---

Why This Matters

Problem	Solution
Auto-combos always gave equal weight (1) to every model, ignoring capability differences.	Snapshot weights differentiate reasoning/vision-capable models at creation time based on the active weight pack.
No way to "freeze" an auto-combo into a static combo for manual tweaking or versioning.	/api/combos/duplicate materializes any auto/* template as a persistent combo with preserved weights and metadata.

---

API Reference — Duplicate Endpoint

POST /api/combos/duplicate
Authorization: Bearer <management-api-key>   (if requireLogin is true)
Content-Type: application/json

{
  "name": "auto/best-coding",    // required — any auto/* template
  "strategy": "priority"         // optional — defaults to "priority"
}

Response (201):

{
  "id": "...",
  "name": "static-best-coding copy",
  "models": [
    { "model": "anthropic/claude-sonnet-4-20250514", "weight": 35 },
    { "model": "openai/gpt-4.1",                    "weight": 28 }
  ],
  "strategy": "priority",
  "config": {
    "sourceAutoCombo": "auto/best-coding",
    "weightPack": { "taskFit": 0.3, "stability": 0.25, ... }
  }
}

Error codes: 400 (missing/invalid name), 401/403 (auth), 422 (unknown template or no matching models), 500 (server error).

---

Files Changed (51 files, +1412 / −249)

Area	Files
Core logic	open-sse/services/autoCombo/virtualFactory.ts
API route	src/app/api/combos/duplicate/route.ts (new)
Dashboard UI	src/app/(dashboard)/dashboard/combos/page.tsx, AutoComboCatalog.tsx
Tests	tests/unit/snapshot-weights.test.ts (new), tests/unit/combos-duplicate-route.test.ts (new), tests/unit/combos-duplicate-resolution-audit.test.ts (new)
i18n	~47 locale files in src/i18n/messages/
Scripts	scripts/ad-hoc/dump-auto-combos.ts (new)

---

Testing

# Snapshot weights
node --import tsx/esm --test tests/unit/snapshot-weights.test.ts

# Duplicate route
node --import tsx/esm --test tests/unit/combos-duplicate-route.test.ts

# All unit tests
npm run test:unit

---

Notes for Reviewers

NOTE! Prettifier page.tsx impact there are too many changes in page.tsx because Prettifier made a bunch of formatting changes. Main change is the OnComboCreated handler passed to AutoComboCatalog! Confirmed with AI that the large page.tsx changes are purely formatting.

1. Snapshot weights are static — they use capabilities and tier info available at combo creation time, not runtime telemetry (p95 latency, quota remaining). This is intentional; the function docstring explains the trade-off.
2. Weight normalization: snapshot scores are clamped to [0, 1], then normalized so all model weights sum exactly to 100 in the duplicate endpoint (with remainder distributed round-robin).
3. Error sanitization follows Rule fix(ui): fix Select dropdown dark theme inconsistency #12 — no raw stack traces in responses.

[swingtempo]

@diegosouzapw
Need confirmation from product team that this is desired and is a bug that was fixed. This isn't an area I'm familiar with.

[swingtempo]

@diegosouzapw These huge sections here are due to reformatting from Prettifier. I think this is automatic.

[diegosouzapw]

Deferido para revalidação do head atual. A rodada anterior registrou um hold de ownership, mas a inspeção atual não confirmou esse worktree; por favor, reconfirme o head e os checks antes da decisão final.

[diegosouzapw]

Babysit summary (Unit Tests fast-path 4/4)

Root cause (from run 31853607931, job Unit Tests fast-path (4/4), tests/unit/i18n-glossary-consistency-check.test.ts + tests/unit/api-route-body-validation.test.ts / t06 gate):

1. POST /api/combos/duplicate (src/app/api/combos/duplicate/route.ts) read request.json() and did hand-rolled typeof checks instead of routing through validateBody()/Zod — fails the t06 "every API route reading request.json() validates it with Zod" gate (Hard Rule chore(env): sync .env.example with current .env structure #7).
2. duplicateAutoComboSnapshotMsg in src/i18n/messages/zh-CN.json used the retired glossary synonym 提供商 instead of the canonical 提供者 for "provider" — fails the i18n glossary consistency regression guard (post-feat(i18n): add native-quality zh-CN terminology and visual QA gates #8038 normalization).

Fix (commit 4f911f6, authored by @swingtempo, co-authored by @diegosouzapw):

• Added duplicateAutoComboSchema (src/shared/validation/schemas/combo.ts) and routed the route body through validateBody()/isValidationFailure() (src/shared/validation/helpers.ts), preserving the same error message/shape for existing tests.
• Swapped 提供商 → 提供者 in the zh-CN string.

Verification:

• node --import tsx/esm --test tests/unit/combos-duplicate-route.test.ts — 10/10 pass.
• node --import tsx/esm --test tests/unit/combos-duplicate-resolution-audit.test.ts — 1/1 pass.
• npx eslint src/app/api/combos/duplicate/route.ts src/shared/validation/schemas/combo.ts — clean.
• Pushed to swingtempo/duplicate-auto-combo2; fresh CI run triggered (run 32438457440 at time of writing).

Threads left unresolved (not addressed by this fix, out of scope):

• virtualFactory.ts — product-confirmation question, unrelated to CI failure.
• dashboard/combos/page.tsx — Prettier reformatting note, unrelated to CI failure.

dast-smoke was previously cancelled/showed fail after a 30m timeout window in the prior run — no assertion failure observed; re-run triggered along with the rest of the fresh CI run. Not modified.

Ready for CI to finish + human review/merge.

[diegosouzapw]

⚠️ base-red inherited: #9985

New CI run (32438457440, after the fix push) shows 2 additional reds unrelated to this PR's diff — both are inherited base drift, not caused by this branch:

• Docs Gates (fast-path): provider-count drift (docs say 346, live code now defines 347) + stale SVG diagrams — a base-tip content change, not touched by this PR.
• No new ESLint warnings: eslintErrors: 0, eslintWarnings: 0 in the job's own metrics — the failure is the quality-ratchet step (openapiCoverage.pct: 38.6 < baseline 39.2), also base-tip drift per the open base-red issue 🔴 Release branch not green: release/v3.8.50 #9985 ("ratchet drift ... is expected mid-cycle and is rebaselined at release — it is NOT a contributor concern").

Original target ("Unit Tests fast-path (4/4)") is fixed by commit 4f911f6 — see prior comment. Not chasing the base-red items per project policy (Hard Rule #21 / base-green protocol); they'll clear on the next base merge or release rebaseline.