[obvious-autobuild]

fleet-packet: burst10-fl-lechef-socket-green

Human author: Sam Gu (sgu07966@gmail.com). This PR (opened as #186, branch obv/burst9-fridgeluck-lechef-socket) carries burst9's phone–gateway E2E work; the burst10 continuation fixed the goal-path CI blocker and re-scoped the branch green.

What changed and why

The feature (burst9, unchanged by this continuation)

The phone and the LeChef gateway had never been exercised together: the gateway's tests fake the model side, and the phone's live stack was only checked by iOS compile/tests. This PR adds the missing seam-to-seam walk — the pinned production phone sources speak the real WebSocket protocol to the real gateway, whose model side is a scripted fake.

• refactor(backend) (676d7e8): server.ts keeps only production wiring; the HTTP routes and live-session gateway move to injectable agentServer.ts so an in-process boot can mount them with fakes. The gateway's frame holder gains an injectable default. No production behavior change.
• test(e2e) (c59bcbe): apps/ios/Tools/lechef-e2e builds the pinned production phone sources (WebSocket client, proposal coordinator, wire codec, proposal executor, CookMode types) on Linux with the documented rewrites (FoundationNetworking, CryptoKit, Combine @Published, excluded session-context methods) and a walk-kitchen executor implementation. src/lechef/e2e/boot.ts mounts the parameterized server with an in-memory store, the scripted fake live session, and an observable frame holder, then spawns the walk and asserts the gateway side.

The session the walk runs: bounded snapshot (revision 7, pantry, recipe, eaters) → shutter frame → timer proposal proposed/approved/executed → butter swap to soy sauce proposed and rejected by the allergen re-check (Ana, wheat) with the contract-shaped recheck receipt → recipe finished on the device → re-snapshot → getStepContext answered from the phone's state (done, step 4) → stale step advance refused ("no active step") → close; session doc deleted and frame cache evicted.

Burst10 delta: the goal-path blocker fixed, the branch re-scoped green

• Original blocker (head 6bd3abb, CI run 38851697494): GoalPathTests.testFridgeToMealToTasteGoalPath failed at lines 295 (stock 1380→1380, "logging the meal consumed pantry stock") and 323 (learned-layer evidence nil). Root cause: a merge-timing race, not product code — merge 6851bbb (04:29:03) pulled obv/burst8-fridgeluck-goal-path one commit before e87ee7b (04:39:08) replaced the strict test there with a pin of actual behavior ("catalog cooks deduct no pantry stock yet — known follow-up"; "a 4-star cook contributes availability, not a low-rating numerator … by design"). test(ios): wire the overhaul into the app and prove the fridge-to-meal-to-taste path #182/feat(evaluation): re-gate taste residual on the app recommendation pool #184/feat(ios): serve catalog recipes to the assistant, finalization, and cloud ranking #185 pass because they carry e87ee7b. No product commit anywhere implements catalog stock deduction (git log -S inventoryConsumption|applyConsumption across all refs).
• Fix: merged the goal-path tip (merge aea3b8f) so the branch carries the canonical test blob 7648a9c0 — byte-identical to the version passing CI on test(ios): wire the overhaul into the app and prove the fridge-to-meal-to-taste path #182/feat(evaluation): re-gate taste residual on the app recommendation pool #184/feat(ios): serve catalog recipes to the assistant, finalization, and cloud ranking #185. This branch authored neither test version; nothing was weakened here.
• The previous continuation had also merged the safety-gates tip (ab6c9f1). CI run 38856495680 on that head showed the goal-path blocker gone (0 GoalPath errors) but the safety-gates feature's own suite failing — 934 sweep assertion failures (testUnknownClassMutationRemovesRecipe 473, testEveryRestrictedHouseholdSweep 458, 3 more cases) plus SubstitutionServiceTests.testExplicitAllergenBlocksTheSubstituteEvenWithoutADiet. The same signature fails on the rejoin line's repaired head bc155f2 (run 38855840833): the breakage is base-borne on obv/burst9-fridgeluck-safety-gates and is owned by PRs feat(ios): one allergen and diet screen behind every offering surface #187/feat(release): the release-candidate build with every late fix, the goal path and the safety sweep green #188, which are actively repairing it.
• Revert commit 43ae74d (git revert -m 1 ab6c9f18, normal fast-forward push ab6c9f1..43ae74d, no force-push) drops that speculative merge: this branch is the LeChef socket E2E work + the alpha merge + the goal-path pin. Post-revert tree is byte-identical to aea3b8f (git diff aea3b8fe HEAD — empty). The safety-gates work returns once feat(ios): one allergen and diet screen behind every offering surface #187/feat(release): the release-candidate build with every late fix, the goal path and the safety sweep green #188 land and re-merge.

How it was checked (commands and results)

Post-revert head 43ae74d:

• Hosted macOS iOS CI: run 38857266926 — build-and-test PASS (5m55s), dead-code-advisory PASS (3m23s). This is the PR's required check, green on the pushed head.
• Portable backend suites re-run in a fresh clone at 43ae74d: bun install (222 pkgs); bun run check — tsc clean; bun test src/__tests__ — 338 pass, 0 fail (5006 expect() calls, 23 files); bun run src/run/lechef-suite.ts — 65/65 job successes, 0 unsafe answers, 0 unauthorized actions, gates all true.
• GoalPathTests blob re-verified on the pushed ref: git show origin/obv/burst9-fridgeluck-lechef-socket:apps/ios/Tests/GoalPath/GoalPathTests.swift | git hash-object --stdin = 7648a9c0 (the test(ios): wire the overhaul into the app and prove the fridge-to-meal-to-taste path #182/feat(evaluation): re-gate taste residual on the app recommendation pool #184/feat(ios): serve catalog recipes to the assistant, finalization, and cloud ranking #185 version).

Intermediate heads (evidence trail):

• 6bd3abb, run 38851697494: failed only GoalPathTests.testFridgeToMealToTasteGoalPath (the two asserts above) — exact failure inventory from --log-failed.
• ab6c9f1, run 38856495680: GoalPath errors 0; SafetyGatesSweepTests + 1 substitution case failing (counts above) — the reason for the revert.
• Feature verification at 6bd3abb (burst9 continuation): ./apps/ios/Tools/lechef-e2e/run.sh walk 11/11 + boot 12/12 ("phone and gateway ran the cook session end to end ✓"); bun test src/__tests__ 338 pass / 0 fail; lechef suite 65/65; bun run check clean; swiftc -parse apps/ios/Tests/LeChefPlanWeeklyTests.swift clean; bunx oxlint 0 errors, 1 pre-existing warning (ToolResponsePayload, liveSessionGateway.ts:181).

What is not verified

• No Xcode/macOS build in this sandbox — the Linux walk compiles the pinned sources, but the iOS app itself is verified only by the hosted macOS CI run cited above.
• Tools/lechef-e2e is a standalone SPM package not referenced by project.yml, so CI does not build or run it; the walk was proven locally at 6bd3abb and its sources are untouched by the burst10 delta.
• The model side is a scripted fake: no real Gemini call was made; prompt/model behavior is out of scope.
• The safety-gates suite is not exercised on this branch after the revert (the feature left with the revert); its green state will be re-established via feat(ios): one allergen and diet screen behind every offering surface #187/feat(release): the release-candidate build with every late fix, the goal path and the safety sweep green #188.

Assumptions

• The goal-path strict test (548f3f4) encoded a future product state; e87ee7b — the goal-path branch's own head, green on test(ios): wire the overhaul into the app and prove the fridge-to-meal-to-taste path #182/feat(evaluation): re-gate taste residual on the app recommendation pool #184/feat(ios): serve catalog recipes to the assistant, finalization, and cloud ranking #185 — is the canonical current spec for that test. Syncing this branch to it resolves a merge-timing artifact; it is not a test weakening.
• "Catalog-cooked meals do not yet deduct pantry stock" remains the product truth (tracked feature-sized follow-up); the strict test's pantry expectation belongs to that future feature, not to this PR.
• The safety-gates sweep repair is feat(ios): one allergen and diet screen behind every offering surface #187/feat(release): the release-candidate build with every late fix, the goal path and the safety sweep green #188's scope; reverting the speculative merge keeps this PR focused. A later merge of the repaired line can bring it back green.
• No merge, no force-push, no base-branch push occurred: the revert is a normal child commit (fast-forward ab6c9f1..43ae74d).

🔗 Obvious Project · 🧵 Obvious Thread · 🔀 Obvious PR Artifact

[coderabbitai]

Important

Review skipped

Bot user detected.

To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 8f5241d4-ba21-41b8-b5f2-12b7d2e8e231

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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

[codemetrics-ai]

Note

Review skipped

Codemetrics is installed on this repository, but it hasn't been added to your team yet.

To enable AI code reviews for this repo, go to your dashboard and add FridgeLuck to your team's repositories.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-16T07:36:32.262554Z	5da8444	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.