Skip to content

work-queue.md

Latest commit

 

History

History
391 lines (353 loc) · 43.3 KB

work-queue.md

File metadata and controls

391 lines (353 loc) · 43.3 KB
Raw
type Reference
title Monorepo tooling rollout work queue
description Ephemeral phase tracker for rolling out Nx local cache, deterministic prepare graph, declaration maps, dependency-cycle linting, and dev watch in the React Native Firebase monorepo.
tags
monorepo
tooling
nx
work-queue
rollout
timestamp 2026-07-10 00:00:00 UTC

Monorepo tooling — rollout work queue

COMPLETE (2026-07-11): MT0.0–MT4 + docs committed. MTV closed — validation + review GREEN; branch ready for PR. Next pickup: MT-WATCH (deferred) or PR. Goal/order: update Lerna → guardrails first (cycle lint, docs, benchmark) → deterministic cached prepare (graph + nx.json incl. complete outputs + scoped inputs + no-cloud + ai split) → declaration maps → dependency-rule hardening → full validation. Dev watch / e2e-rerun is deferred to a later gap-analysis pre-phase (not on the critical path).

Ephemeral tracker; see OKF policy. Work types / tiers / gate field ids: iteration vocabulary. Loop, gates, host rule, harness: change authoring workflow — not restated. Agent commands: agent command policy only — no yarn workspace … prepare, no Jet probes.

Durable decisions: architecture-decisions.md. Design detail: prepare-and-cache.md.

Locked decisions (index)

ADR Decision
MonoTool-AD-1 Nx local cache via Lerna runner; no Turborepo; no Nx Cloud
MonoTool-AD-2 Keep Lerna for versioning + publish
MonoTool-AD-3 Build order via devDependencies
MonoTool-AD-4 One inlined prepare command; no wrapper script
MonoTool-AD-5 Keep bob; emit declaration maps
MonoTool-AD-6 dependency-cruiser as lint:deps
MonoTool-AD-7 AI fixtures = Jest prerequisite, not build step
MonoTool-AD-8 .nx/cache on CI (with NX_NO_CLOUD), not on publish
MonoTool-AD-9 Dev watch + e2e rerun off Metro — Deferred (gap-analysis pre-phase)
MonoTool-AD-10 Generated version files are declared cache outputs, not committed
MonoTool-AD-11 Scope prepare cache inputs via a jsSource namedInput
MonoTool-AD-12 Never Nx-cache prepare when the script is patch-package

Phase ordering

Phase Scope Depends on Why
MT0.0 Update lerna to current (pulls current bundled nx) — Prereq: nx.json keys + nx watch fixes; cheap
MT0.1 dependency-cruiser lint:deps (no-circular) + scoped config + CI + change-authoring gate MT0.0 Guardrail; cheap; catches cycles before graph edits
MT0.2 Durable OKF docs (this bundle) — Landed (planning + adversarial-review revision)
MT0.3 scripts/benchmark-prepare.sh + pre-Nx baseline numbers MT0.0 Baseline before cache claims
MT1 devDependency graph (17 pkgs) + nx.json (scoped inputs + complete outputs + no-cloud) + inline prepare (name unchanged) + ai prepare split + gated AI fetch + CI .nx/cache (NX_NO_CLOUD) MT0.3 Deterministic, cached prepare
MT2 declarationMap in base + ai + vertexai tsconfig MT1 IDE go-to-definition; cheap follow-on
MT4 dependency-cruiser rule hardening (not-to-own-dist scoped + hub/chain allowlist) MT1 Enforce graph after devDeps land
MT-WATCH Deferred — gap-analysis pre-phase: dev watch + event-driven e2e rerun MT1 Not built today; needs detailed analysis before implementation; off critical path
MTV Branch-wide validation MT1, MT2, MT4 Merge gate (MT-WATCH not required)

Resume checklist

Gate prerequisites before any :test-cover (host rule):

  1. Pre-flight: host-clear probes, services ready, harness matches validation tier; serial :test-cover runs; frozen tree for independent-review.
  2. Most items here are config/build/docs (no :test-cover); e2e applies only where the row's validation_tier names it (MTV, and the deferred MT-WATCH if/when it reaches implementation).

Per-item detail

Each item is one serial loop: implementation (unit-focused) → independent-review (frozen) → commit. Gate fields: iteration vocabulary. No item touches packages/*/lib/** runtime or android/ios native, so coverage_evidence_gate: n/a throughout (verify per item against the coverage grep).

MT0.0 — Update Lerna to current (prerequisite)

  • next_work_type: commit · validation_tier: area-focused · gates: impl closed, review closed, commit closed · commit_subject: build(deps): refresh nx for lerna runner
  • Do:
    • Bump lerna to the latest release in root devDependencies; refresh yarn.lock. This pulls a current bundled nx (MonoTool-AD-1 prerequisite).
    • Confirm the bundled nx version supports the nx.json keys used in MT1 (neverConnectToCloud, namedInputs, per-target cache/inputs/outputs) and includes the nx watch --all --initialRun fix (nx PR #32282) needed by the deferred MT-WATCH.
  • Acceptance:
    • yarn installs cleanly; yarn lerna:prepare still exits 0 (behavior unchanged pre-nx.json).
    • yarn lint, yarn tsc:compile, yarn tests:jest remain green.
    • Record the resulting lerna + bundled nx versions in a note (feeds MT1/MT-WATCH version checks).
  • Implementation evidence (unit-focused, 2026-07-10):
    • Change: lerna was already current at 9.0.7; refreshed transitive nx / @nx/devkit from 22.1.3 → 22.7.6 via lockfile refresh (no root resolutions pins; Lerna's >=21.5.3 <23.0.0 range resolves to latest 22.x).
    • Files changed: yarn.lock (Nx-family bump only; no package.json pin).
    • Version evidence: lerna 9.0.7 → 9.0.7; bundled nx 22.1.3 → 22.7.6; nx-schema.json includes neverConnectToCloud, namedInputs, and per-target cache/inputs/outputs.
    • Validation evidence:
      Command Exit Notes
      yarn 0 Clean install; peer warnings only
      yarn lerna:prepare 0 20 projects prepared successfully
      yarn lint 0 JS + Android + iOS lint/checks green
      yarn tsc:compile 0 Root TS compile clean
      yarn tests:jest 0 82 suites / 1172 tests passed; log /tmp/mt0-jest.log
  • Independent-review findings (area-focused, 2026-07-10):
    • Serious (initial): first pass used resolutions for nx only; @nx/devkit skewed at 22.1.3. Remediation replaced pins with lockfile refresh (no root resolutions); Nx-family aligns at 22.7.6.
    • Minor: commit subject updated to build(deps): refresh nx for lerna runner.
    • Review validation reran green before finding: yarn 0 (/tmp/mt0-review-yarn.log), yarn lerna:prepare 0, yarn lint 0 (/tmp/mt0-review-lint.log), yarn tsc:compile 0 (/tmp/mt0-review-tsc.log), yarn tsc:compile:consumer 0 (/tmp/mt0-review-tsc-consumer.log), yarn tests:jest 0 (/tmp/mt0-review-jest.log, 82 suites / 1172 tests). Coverage evidence: n/a (no packages/*/lib/** or native bridge sources).
  • Remediation evidence (unit-focused, 2026-07-10):
    • Change: removed nx / @nx/devkit resolutions; yarn install lockfile refresh aligns all Nx-family packages at 22.7.6 (no 22.1.3 entries remain).
    • Version evidence: lerna 9.0.7 unchanged; nx 22.7.6; @nx/devkit 22.7.6; @nx/nx-darwin-arm64 22.7.6; nx-schema.json keys present; nx watch --all --initialRun supported at 22.7.6 (>= 22.6.0 threshold).
    • Validation evidence:
      Command Exit Notes
      yarn 0 Clean install; peer warnings only; log /tmp/mt0-remediation-yarn.log
      yarn lerna:prepare 0 20 projects prepared successfully; log /tmp/mt0-remediation-lerna-prepare.log
      yarn lint 0 JS + Android + iOS lint/checks green; log /tmp/mt0-remediation-lint.log
      yarn tsc:compile 0 Root TS compile clean; log /tmp/mt0-remediation-tsc.log
      yarn tests:jest 0 82 suites / 1172 tests passed; log /tmp/mt0-remediation-jest.log
      yarn tsc:compile:consumer 0 Consumer TS compile clean; log /tmp/mt0-remediation-tsc-consumer.log
  • Independent-review evidence after remediation (area-focused, 2026-07-10):
    • Findings: none. Serious @nx/devkit skew fixed; no 22.1.3 Nx-family entries remain in yarn.lock; nx, @nx/devkit, and all @nx/nx-* platform packages resolve to 22.7.6.
    • Diff scope verified: package.json unchanged by Nx pinning; lerna remains ^9.0.7 / CLI 9.0.7; lockfile churn is Nx-family transitive updates, not unrelated scope.
    • Validation evidence:
      Command Exit Notes
      yarn 0 Clean install + postinstall prepare; log /tmp/mt0-fresh-review-yarn.log
      yarn lerna:prepare 0 20 projects prepared
      yarn lint 0 JS + Android + iOS checks green; log /tmp/mt0-fresh-review-lint.log
      yarn tsc:compile 0 Root TS compile; log /tmp/mt0-fresh-review-tsc.log
      yarn tests:jest 0 82 suites / 1172 tests; log /tmp/mt0-fresh-review-jest.log
      yarn tsc:compile:consumer 0 Consumer TS compile; log /tmp/mt0-fresh-review-tsc-consumer.log
    • Coverage evidence: n/a (no packages/*/lib/** or native bridge sources).

MT0.1 — dependency-cruiser lint:deps

  • next_work_type: commit · validation_tier: area-focused · gates: impl closed, review closed, commit closed · commit_subject: build(deps): add dependency-cruiser lint:deps for lib import graph
  • Do:
    • Add dependency-cruiser devDependency (root).
    • Add .dependency-cruiser.cjs with rule no-circular on packages/*/lib/** (allowlist + not-to-own-dist hardening deferred to MT4). Scope the config now (design): doNotFollow/exclude for node_modules, packages/*/dist/**, packages/*/__tests__/**, packages/*/e2e/**; set options.tsConfig + enhancedResolveOptions so the @react-native-firebase/* path aliases resolve.
    • package.json: "lint:deps": "depcruise --config .dependency-cruiser.cjs packages"; add yarn lint:deps into the JS lint path.
    • CI: add a lint:deps step in .github/workflows/linting.yml lint job.
    • okf-bundle/testing/validation-checklist.md § lint + change-authoring-workflow.md lint gate: note lint:deps blocking when packages/*/lib/** in diff (durable OKF edit → DRY pass).
  • Acceptance:
    • yarn lint:deps exits 0 on current tree (graph is a DAG today).
    • Resolution sanity: at least one real cross-package edge (e.g. a satellite → @react-native-firebase/app/...) is resolved, not reported as unresolvable (proves the tsconfig-alias wiring works).
    • yarn lint (full) still exits 0.
    • Introducing a deliberate temporary cycle makes yarn lint:deps exit non-zero (then revert).
    • CI lint job runs the step.
  • Implementation evidence (unit-focused, 2026-07-10):
    • Change: added dependency-cruiser@17.4.3; new .dependency-cruiser.cjs with scoped inter-package no-circular on packages/*/lib/**, excludes for node_modules/dist/__tests__/e2e, TS path-alias resolution via generated tsconfig.depcruise.json; root lint:deps wired into yarn lint; CI step in .github/workflows/linting.yml; OKF validation gates updated in validation-checklist.md + change-authoring-workflow.md.
    • Files changed: package.json, yarn.lock, .dependency-cruiser.cjs, .github/workflows/linting.yml, okf-bundle/testing/validation-checklist.md, okf-bundle/testing/change-authoring-workflow.md.
    • Validation evidence:
      Command Exit Notes
      yarn lint:deps 0 32 modules / 74 deps; log /tmp/mt01-lint-deps.log
      yarn lint 0 JS + deps + Android + iOS; log /tmp/mt01-lint.log
      temporary cycle → yarn lint:deps 3 app ↔ crashlytics cycle detected; log /tmp/mt01-lint-deps-cycle.log
      post-revert yarn lint:deps 0 temp import reverted; clean tree restored
    • Resolution sanity: packages/analytics/ → packages/app/ edge resolved (@react-native-firebase/app/dist/module/common, types aliased-tsconfig-paths); 0 unresolved from analytics; log /tmp/mt01-resolution-sanity.log.
    • Follow-ups flagged by implementer: generated tsconfig.depcruise.json each run (consider .gitignore); graph currently analyzes compiled lib/**/*.js (excludes .ts); CI runs lint:deps twice via dedicated step + yarn lint.
  • Independent-review findings (area-focused, 2026-07-10):
    • Serious: config excludes authored lib/**/*.ts and effectively analyzes ephemeral/committed .js sidecars, not the real TS import graph; on a clean CI tree without local compile sidecars, MT0.1 acceptance is not proven. Remediation required: remove .ts exclude, cruise authored TS under packages/*/lib/**, and re-validate on a clean tree.
    • Minor: gitignore generated tsconfig.depcruise.json; tighten cruise scope/excludes (plugin/__tests__, non-lib/** noise); deduplicate CI lint:deps invocation (dedicated step or yarn lint, not both).
    • Review validation: yarn lint:deps 0 (/tmp/mt01-review-lint-deps.log); yarn lint/yarn lint:js failed on this host due to ephemeral untracked packages/*/lib/**/*.js sidecars (~19.5k eslint errors), outside MT0.1 frozen scope but relevant to clean-tree proof. Coverage evidence: n/a.
  • Remediation evidence (unit-focused, 2026-07-10):
    • Change: TS-first depcruise config (includeOnly on packages/*/lib/**/*.ts(x), moduleResolution: 'node', noEmit: true); scoped excludes; gitignored tsconfig.depcruise.json; lint:deps scoped to packages/*/lib; CI deduped to single yarn lint step.
    • Files changed: .dependency-cruiser.cjs, .gitignore, package.json, .github/workflows/linting.yml, okf-bundle/testing/validation-checklist.md.
    • Validation evidence:
      Command Exit Notes
      yarn lint:deps (clean TS tree) 0 19 modules / 23 deps; log /tmp/mt01-remediation-lint-deps-clean.log
      resolution sanity (analytics → app) 0 @react-native-firebase/app → packages/app/ via aliased-tsconfig-paths; log /tmp/mt01-remediation-resolution-sanity.log
      temp cycle → yarn lint:deps 3 app ↔ crashlytics; log /tmp/mt01-remediation-lint-deps-cycle.log; reverted
      post-revert yarn lint:deps 0 log /tmp/mt01-remediation-lint-deps-post-revert.log
      yarn lint:js 0 ESLint green
      yarn lint 1 lint:deps passes; lint:android failed (google-java-format exit 1) — implementer claims pre-existing Android formatter flake unrelated to MT0.1
    • Clean-tree proof: validation used authored TS only (ephemeral untracked lib/**/*.js sidecars temporarily moved aside; tracked memidb .js retained).
  • Independent-review evidence after remediation (area-focused, 2026-07-10):
    • Findings: no serious findings. One minor local-only caveat: with ephemeral untracked lib/**/*.js sidecars present, standalone yarn lint:deps can vacuously pass (0 deps cruised); mitigated because yarn lint runs lint:js first and CI uses clean checkout.
    • Validation evidence (clean TS tree):
      Command Exit Notes
      yarn lint:deps 0 19 modules / 23 deps cruised; log /tmp/mt01-fresh-review-lint-deps-clean.log
      resolution sanity (analytics → app) 0 @react-native-firebase/app → packages/app/
      yarn lint:js 0 log /tmp/mt01-fresh-review-lint-js
      yarn lint 0 js + deps + android + ios; log /tmp/mt01-fresh-review-lint-clean.log
      yarn lint:android (isolated) 0 prior android flake not reproduced
    • Coverage evidence: n/a.

MT0.2 — Durable OKF docs — landed

  • work type: documentation · validation_tier: none · gates: n/a · commit_subject: docs: add monorepo tooling OKF bundle
  • Do: okf-bundle/monorepo-tooling/{index.md,architecture-decisions.md,prepare-and-cache.md,work-queue.md}; add a "Monorepo tooling" entry to okf-bundle/index.md.
  • Acceptance: OKF DRY/consistency pass green (canonical location, link hygiene, durable vs ephemeral split); decisions live only in architecture-decisions.md; ephemeral phases live only here. Documentation/DRY pass completed after docs: correct assertion about JS bundles in dev.

MT0.3 — Benchmark baseline

  • next_work_type: commit · validation_tier: none · gates: impl closed, review closed, commit closed · commit_subject: build(scripts): add prepare benchmark script
  • Do: scripts/benchmark-prepare.sh (macOS; scenarios A–D from prepare-and-cache § benchmark); capture pre-Nx B/C/D medians into a benchmarks note.
  • Acceptance:
    • Script runs A–D, prints median-of-3 per scenario, exits 0.
    • Pre-Nx baseline numbers recorded (used to prove MT1 gains).
  • Implementation evidence (none tier, 2026-07-10):
    • Change: added scripts/benchmark-prepare.sh (macOS guard, scenarios A–D, median-of-3, reversible firestore edit with EXIT trap); ephemeral baseline note prepare-benchmark-baseline.md.
    • Files changed: scripts/benchmark-prepare.sh, okf-bundle/monorepo-tooling/prepare-benchmark-baseline.md.
    • Pre-Nx medians (s): A 125.583; B 23.686; C 23.694; D 23.758. Logs: .tmp/prepare-benchmarks/20260710-122118/.
    • Validation evidence:
      Command Exit Notes
      bash ./scripts/benchmark-prepare.sh 0 Full A–D run (~624s unsandboxed); sandboxed first attempt failed on partial node_modules removal
      bash -n scripts/benchmark-prepare.sh 0 Syntax check
    • Coverage evidence: n/a (no packages/*/lib/** runtime or native bridge edits committed).
  • Independent-review evidence (none tier, 2026-07-10):
    • Findings: none. Script aligns with prepare-and-cache.md benchmark methodology; baseline numbers live in ephemeral note only.
    • Validation evidence:
      Command Exit Notes
      bash -n scripts/benchmark-prepare.sh 0 Review smoke syntax check
    • Coverage evidence: n/a.

MT1 — Deterministic cached prepare

  • next_work_type: commit · validation_tier: area-focused · gates: impl closed, review closed, commit closed · commit_subject: build(deps): enforce package build order and enable nx prepare cache
  • Do:
    • Add "@react-native-firebase/app": "<version>" to devDependencies of the 17 packages = every package except app and vertexai (16 plain satellites + ai); vertexai orders transitively via its ai dependency (MonoTool-AD-3; counts owned by prepare-and-cache § graph).
    • Add nx.json — design: neverConnectToCloud: true; namedInputs.jsSource; prepare target dependsOn: ["^prepare"], inputs: ["jsSource"] (MonoTool-AD-11), cache: true, outputs = dist/** + plugin/build/** + lib/version.ts (MonoTool-AD-10).
    • Add an app-scoped nx.targets.prepare.outputs override in packages/app/package.json re-listing dist/**, plugin/build/**, lib/version.ts, ios/RNFBApp/RNFBVersion.m, android/src/reactnative/java/io/invertase/firebase/app/ReactNativeFirebaseVersion.java (MonoTool-AD-10).
    • Update the yarn lerna:prepare body (name unchanged) to cross-env NX_NO_CLOUD=true lerna run prepare (MonoTool-AD-4).
    • packages/ai: drop yarn tests:ai:mocks from prepare; add an AI-test-gated Jest prerequisite (single flat jest config — a globalSetup that no-ops unless the run includes packages/ai/__tests__/**, or a projects split) running the fetch + convert scripts; make the fetch offline-idempotent (local-clone short-circuit before git ls-remote) (MonoTool-AD-7).
    • CI: restore/save .nx/cache via actions/cache in PR/main jobs; set NX_NO_CLOUD=true on every Nx-running job; do not touch publish.yml (MonoTool-AD-8).
    • Remove scoped-prepare guidance from validation-checklist.md (durable OKF edit → DRY pass).
  • Acceptance:
    • Determinism: from a clean dist (rm -rf packages/*/dist packages/*/plugin/build), yarn lerna:prepare exits 0 repeatably (run ≥3×); packages/app/dist/** exists before any satellite compile (no ordering race). If ordering is not honored, switch the inline body to nx run-many -t prepare (MonoTool-AD-4 fallback).
    • Cache-replay completeness (regression guard for MonoTool-AD-10): after a warm prepare, delete the generated files (git clean -fdx -- 'packages/*/lib/version.ts' packages/app/ios/RNFBApp/RNFBVersion.m 'packages/app/android/**/ReactNativeFirebaseVersion.java') and packages/*/dist, then re-run yarn lerna:prepare (cache hit) → all generated files are restored, and yarn tests:jest + yarn tsc:compile still exit 0. (Simulates the fresh-CI-checkout + restored-.nx/cache path.)
    • yarn tsc:compile and yarn tsc:compile:consumer exit 0.
    • yarn tests:jest exits 0, including packages/ai (fixtures fetched via the gated prerequisite, not prepare); a non-AI yarn tests:jest run performs no network fetch.
    • Input scoping (MonoTool-AD-11): editing a package's __tests__/e2e/android/ios file does not invalidate its prepare cache; editing its lib/** does.
    • A no-op yarn lerna:prepare (MT0.3 scenario C) is materially faster than the pre-Nx baseline (cache hit); single-package edit (D) rebuilds only that package + dependents.
    • Publish-flow check (MonoTool-AD-2): a lerna version dry run rewrites the new internal @react-native-firebase/app devDep ranges to the bumped version and shows no other release-flow change (no side effects).
    • git grep vertexai-sdk-test-data shows fixtures still gitignored; yarn (fresh) does not clone test data.
    • Optional (review discretion): one platform e2e smoke to confirm dist/module/** output is unchanged by the ordering change.
  • Implementation evidence (unit-focused, 2026-07-10):
    • Change: 17-package @react-native-firebase/app devDep graph; root nx.json (scoped inputs/outputs, no-cloud); app nx output override; cross-env NX_NO_CLOUD=true lerna run prepare; AI mocks out of prepare into gated Jest globalSetup; fetch script local-clone fast path; CI .nx/cache restore/save + NX_NO_CLOUD on jest/lint/e2e jobs; validation-checklist scoped-prepare line removed.
    • Files changed: nx.json, package.json, packages/{app,ai,16 satellites}/package.json, jest.config.js, scripts/jest-ai-mocks-global-setup.js, scripts/fetch_ai_mock_responses.ts, .github/workflows/{tests_jest,linting,tests_e2e_*}.yml, okf-bundle/testing/validation-checklist.md, yarn.lock.
    • Validation evidence:
      Command Exit Notes
      yarn 0 Fresh install + postinstallDev prepare (20 projects); /tmp/mt1-yarn-fresh.log
      yarn lerna:prepare ×3 0 Warm cache ~1.38–1.40s each; /tmp/mt1-lerna-prepare-repeat.log
      Determinism (clean dist ×3) 0 packages/app/dist/module/index.js present; /tmp/mt1-determinism.log
      Cache-replay (AD-10) 0 20/20 cache hits; version + native files restored; /tmp/mt1-cache-replay-*.log
      yarn tsc:compile 0 /tmp/mt1-tsc.log
      yarn tsc:compile:consumer 0 /tmp/mt1-tsc-consumer.log
      yarn tests:jest 0 82 suites / 1172 tests; /tmp/mt1-jest.log
      Non-AI jest 0 59 suites / 847 tests, no network; /tmp/mt1-jest-non-ai.log
      yarn lint:js + yarn lint:deps 0 /tmp/mt1-lint-{js,deps}.log
      Benchmark (post-Nx A–D) 0 See prepare-benchmark-baseline.md § Post-Nx; logs .tmp/prepare-benchmarks/20260710-152608/; C 1.410s vs pre-Nx 23.694s (~16.8×); B 1.880s vs 23.686s (~12.6×); D median 1.457s vs 23.758s
      lerna changed + devDep spot-check n/a No --dry-run in Lerna 9; devDeps at 25.1.0 on analytics/ai/firestore; /tmp/mt1-lerna-changed.log
      git grep vertexai-sdk-test-data 0 Fixtures gitignored; /tmp/mt1-git-grep-fixtures.log
    • Coverage evidence: n/a (no packages/*/lib/** or native bridge product edits).
    • Deviations noted: full yarn lint (android/ios) not run; lib-edit cache bust not directly exercised; Nx intermittent flaky-task warning on cache-replay (all exit 0).
  • Independent-review findings (area-focused, 2026-07-10):
    • Serious: nx.json + scripts/jest-ai-mocks-global-setup.js untracked — must stage before commit.
    • Serious: shouldFetchAiMocks() ignores CLI --testPathIgnorePatterns=packages/ai (Jest passes pre-CLI config to globalSetup); non-AI runs can still fetch on fresh hosts. Fix: inspect process.argv for ignore patterns containing packages/ai.
    • Minor: non-AI validation command misleading until #2 fixed.
    • Nit: do not commit .dependency-cruiser.min.cjs; android format warning pre-existing.
    • Review validation: all re-runs green; cache-replay AD-10 verified 20/20 hits. Logs /tmp/mt1-review-*.log.
    • Gate: review_gate stays open — remediation required.
  • Remediation evidence (unit-focused, 2026-07-10):
    • Change: getArgvTestPathIgnorePatterns() merges CLI --testPathIgnorePatterns (= and space forms) with globalConfig.testPathIgnorePatterns; AI ignored via argv skips fetch.
    • Files changed: scripts/jest-ai-mocks-global-setup.js.
    • Validation evidence:
      Command Exit Notes
      yarn tests:jest --testPathIgnorePatterns=packages/ai 0 No fetch output; /tmp/mt1-remediation-ignore-equals.log
      yarn tests:jest --testPathIgnorePatterns packages/ai 0 No fetch output; /tmp/mt1-remediation-ignore-space.log
      yarn tests:jest -- packages/firestore 0 No fetch; /tmp/mt1-remediation-firestore.log
      yarn tests:jest -- packages/ai 0 Fetch invoked; /tmp/mt1-remediation-ai-fetch.log
      node --check scripts/jest-ai-mocks-global-setup.js 0 Syntax
  • Independent-review evidence after remediation (area-focused, 2026-07-10):
    • Findings: none. Prior serious (untracked deliverables — content OK; staged at commit; globalSetup argv gating) resolved.
    • Validation evidence:
      Command Exit Notes
      yarn lerna:prepare 0 20/20 cache hits; /tmp/mt1-delta-review-20260710-125221.log
      Non-AI jest (ignore patterns) 0 Zero fetch with clone moved aside
      yarn tests:jest -- packages/ai 0 Fetch invoked when AI included
      yarn tests:jest (full) 0 82 suites / 1172 tests
      yarn tsc:compile + consumer 0
    • Coverage evidence: n/a.

MT2 — Declaration maps

  • next_work_type: commit · validation_tier: area-focused · gates: impl closed, review closed, commit closed · commit_subject: build(ts): emit declaration maps for package builds
  • Do: add declaration: true + declarationMap: true to tsconfig.packages.base.json, packages/ai/tsconfig.json, packages/vertexai/tsconfig.json (MonoTool-AD-5).
  • Acceptance:
    • After yarn lerna:prepare, packages/*/dist/typescript/**/*.d.ts.map exist.
    • yarn tsc:compile exits 0; yarn lint exits 0.
    • Spot-check: IDE go-to-definition on a cross-package symbol lands on lib/** source (record the check in notes).
    • Map resolves to shipped source: open one emitted .d.ts.map and confirm its sources point at the published lib/*.ts (relative, no absolute path leak) — since files publishes lib, external consumers get go-to-definition too.
    • .d.ts.map do not leak into unintended published paths (bob files/exclude still correct).
  • Implementation evidence (unit-focused, 2026-07-10):
    • Change: added declaration: true + declarationMap: true to tsconfig.packages.base.json, packages/ai/tsconfig.json, packages/vertexai/tsconfig.json (MonoTool-AD-5).
    • Files changed: tsconfig.packages.base.json, packages/ai/tsconfig.json, packages/vertexai/tsconfig.json.
    • Map spot-check: FirebaseApp (analytics/lib/index.ts → @react-native-firebase/app → packages/app/dist/typescript/lib/index.d.ts) go-to-def target packages/app/lib/index.ts; index.d.ts.map has sourceRoot: "", sources: ["../../../lib/index.ts"], 0 absolute-path leaks; 310 .d.ts.map emitted, all under dist/typescript (0 in dist/module).
    • Validation evidence:
      Command Exit Notes
      yarn lerna:prepare 0 20 projects rebuilt (base tsconfig change busted cache); /tmp/mt2-prepare.log
      .d.ts.map count/leak scan 0 310 maps; 0 absolute leaks; 0 outside dist/typescript; /tmp/mt2-dtsmap-count.log, /tmp/mt2-leak-scan.log
      yarn tsc:compile 0 /tmp/mt2-tsc.log
      yarn tsc:compile:consumer 0 /tmp/mt2-tsc-consumer.log
      yarn lint (full) 0 js + deps + android + ios; /tmp/mt2-lint-2.log (android formatter flaked once, clean on re-run)
      yarn lint:deps 0 19 modules / 23 deps; /tmp/mt2-lint-deps.log
      yarn tests:jest 0 82 suites / 1172 tests; /tmp/mt2-jest.log
    • Coverage evidence: n/a (tsconfig/build-config only; no packages/*/lib/** or native bridge edits).
  • Independent-review evidence (area-focused, 2026-07-10):
    • Findings: none. Product diff is three tsconfig files only.
    • Validation evidence:
      Command Exit Notes
      yarn lerna:prepare 0 20/20 cache hits; /tmp/mt2-review-prepare.log
      .d.ts.map count/leak scan 0 310 maps; 0 absolute leaks; 0 in dist/module; /tmp/mt2-review-dtsmap-scan.log
      npm pack --dry-run (analytics) 0 11 maps in tarball under dist/typescript; /tmp/mt2-review-publish-check.log
      yarn tsc:compile + consumer 0 /tmp/mt2-review-tsc.log, /tmp/mt2-review-tsc-consumer.log
      yarn lint (full) 0 js + deps + android + ios; /tmp/mt2-review-lint-full.log
      yarn tests:jest 0 82 suites / 1172 tests; /tmp/mt2-review-jest.log
    • Coverage evidence: n/a.

MT4 — dependency-cruiser rule hardening — gated on MT1

  • next_work_type: commit · validation_tier: area-focused · gates: impl closed, review closed, commit closed · commit_subject: build(deps): enforce package import boundaries
  • Do: extend .dependency-cruiser.cjs (MonoTool-AD-6): not-to-own-dist (forbid only relative ../dist/**/./dist/** imports — not the mapped hub API @react-native-firebase/app/dist/module/..., which is legitimate); graph allowlist (satellites → app only; ai → auth/app-check; vertexai → ai).
  • Acceptance:
    • yarn lint:deps exits 0 on current tree with the stricter rules (the existing @react-native-firebase/app/dist/module/... imports in ~15 satellites must not be flagged).
    • A deliberate off-graph import (e.g. firestore importing auth) fails lint:deps (then revert).
    • A deliberate relative own-../dist import fails lint:deps (then revert).
    • yarn lint full still exits 0.
  • Implementation evidence (unit-focused, 2026-07-10):
    • Change: extended .dependency-cruiser.cjs with not-to-own-dist (module scope, same-package capture), hub-no-internal, satellites-only-hub, ai-graph, vertexai-graph; dynamic tsconfig paths for all non-hub packages; includeOnly widened to retain own-dist edges; collapse removed so module-scope violations survive.
    • Files changed: .dependency-cruiser.cjs.
    • Probe evidence:
      Probe Exit Notes
      Clean yarn lint:deps 0 313 modules / 963 deps; /tmp/mt4-lint-deps-clean.log
      Off-graph (firestore → auth) 1 satellites-only-hub; reverted; /tmp/mt4-probe-off-graph.log
      Own-dist (analytics ../dist/module) 1 not-to-own-dist; reverted; /tmp/mt4-probe-own-dist.log
    • Validation evidence:
      Command Exit Notes
      yarn lint:js 0 /tmp/mt4-lint-js.log
      yarn lint (full) 0 /tmp/mt4-lint-full.log
      yarn tsc:compile 0 /tmp/mt4-tsc-compile.log
      yarn tests:jest 0 82 suites / 1172 tests; /tmp/mt4-tests-jest.log
    • Coverage evidence: n/a (config-only).
  • Independent-review evidence (area-focused, 2026-07-10):
    • Findings: minor (non-blocking, out of MT4 acceptance scope): (1) type-only imports bypass graph allowlist — value imports enforced; matches existing ai/vertexai patterns; (2) relative cross-package ../../pkg/dist/ bypasses allowlist path matching — current tree uses alias imports only; own-dist rule catches same-package relative dist.
    • Validation evidence:
      Command Exit Notes
      yarn lint:deps (clean) 0 313 modules / 963 deps; /tmp/mt4-review-lint-deps-clean.log
      Off-graph probe (firestore → auth) 1 reverted; /tmp/mt4-review-probe-off-graph.log
      Own-dist probe 1 reverted; /tmp/mt4-review-probe-own-dist.log
      yarn lint (full) 0 /tmp/mt4-review-lint-full.log
      yarn tsc:compile 0 /tmp/mt4-review-tsc-compile.log
      yarn tests:jest 0 82 suites / 1172 tests; /tmp/mt4-review-tests-jest.log
    • Coverage evidence: n/a.

MT-WATCH — Dev watch + e2e rerun (gap-analysis pre-phase) — Deferred

  • next_work_type: gap-analysis · validation_tier: area-focused (only if it reaches implementation) · gates: impl open, review open, commit open · commit_subject: n/a until analysis converges
  • Status: Deferred — no incremental dev-watch/hot-reload loop exists today; off the critical path for MT1/MT2/MT4/MTV (MonoTool-AD-9). This pre-phase does the detailed analysis that must precede any implementation.
  • Do (analysis):
    • Resolve the Tier A command caveats (design): verify bundled nx includes the --all --initialRun fix (else first pass via nx run-many -t prepare); use nx run-many -t prepare -p $NX_PROJECT_NAME for multi-project watch batches.
    • Note: the OKF debug/release correction already landed separately as docs: correct assertion about JS bundles in dev; running e2e § Rules #3 is now the canonical owner of that fact. MT-WATCH only needs to analyze the dev-watch and event-driven rerun mechanics.
    • Analyze the event-driven e2e rerun (Metro update-done/onBundleBuilt → Jet POST /rerun on 8091; design Tier C; Jet patch workflow); confirm update-done observability and whether an app-side Jet reload hook is simpler; assess coverage-teardown-handshake bypass in fast-rerun mode.
  • Acceptance (analysis):
    • Written findings: viability, chosen command shapes, and a go/no-go for a follow-on implementation phase (with its own gates). If no-go, close as documentation (record findings) — do not force implementation.
    • dev:watch, if later implemented, is human-only — agent-command-policy.md keeps yarn lerna:prepare canonical.

MTV — Branch-wide validation — closed

  • next_work_type: n/a · validation_tier: full · gates: impl closed, review closed, commit closed · commit_subject: n/a (no product edit)
  • Status (2026-07-11): GREEN — validation (e821793c) + independent review (4361b403); no critical/serious findings.
  • Validation evidence:
    • Static: prepare ×3 (21 cache hits), AD-10 replay, tsc ×2, Jest 82/1172, lint/compare:types/reference:api exit 0 — /tmp/mtv2-full-*.log
    • E2e: macOS 710/0/34, iOS 848/0/85, Android 878/0/54 — /tmp/mtv2-full-e2e-*.log
    • Review: lint:deps, tsc, compare:types, prepare spot-check (20/20 hits) — /tmp/mtv2-review-*.log
  • Review follow-ups (non-blocking, post-merge OK): app/scripts/** in nx cache inputs; lint:deps in agent-command-policy registry; AD-4 scoped-prepare wording; CI cache key nit.

Current gates

Item impl review commit next_work_type
MT0.0 closed closed closed commit
MT0.1 closed closed closed commit
MT0.2 closed n/a closed documentation (landed)
MT0.3 closed closed closed commit
MT1 closed closed closed commit
MT2 closed closed closed commit
MT4 closed closed closed commit
MT-WATCH open open open gap-analysis (deferred)
MTV closed closed closed n/a (merge gate passed)