Files
waggle-os/docs/oss/baseline.md

9.7 KiB
Raw Blame History

Baseline at the start of the conversion

The state of build, typecheck, lint and tests before any scope work. Every later change is compared against this record: a check that passes here must not start failing.

  • Date: 2026-10-15
  • Commit: 2346cf3 on oss/community-edition (the commercial code at b20b138 plus the fixes UF-001, UF-007, UF-008, UF-021 and UF-022)
  • Machine: macOS 26.6.2, Apple silicon (arm64), Node 22.23.2, npm 10.9.8
  • CI: not set up yet. The location of the public repository is still open, so this record stands in for a CI baseline until the pipeline exists (work package 0.2).

Results

Check Command Result
Package build npm run build:packages Pass
Typecheck, agent npx tsc --noEmit --project packages/agent/tsconfig.json Pass
Typecheck, server npx tsc --noEmit --project packages/server/tsconfig.json Pass
Typecheck, app tooling npx tsc --noEmit --project app/tsconfig.json Pass
Typecheck, web cd apps/web && npx tsc --noEmit -p tsconfig.app.json Pass
Web build cd apps/web && npx vite build --outDir <temp dir> Pass
Lint, root npx eslint . (ignores apps/web, which has its own config) 1,516 files, 0 errors, 0 warnings
Lint, web cd apps/web && npx eslint . 605 files, 49 errors (43 no-explicit-any), 60 warnings, all pre-existing
Unit tests, root npm run test 719 files, 12,012 tests: 11,915 passed, 41 failed, 56 skipped (8.6 min)
Unit tests, web cd apps/web && npx vitest run 221 files, 1,979 tests, all passed (2.4 min)
E2E npm run test:e2e Not run, see below
Visual regression npm run test:visual Not run, see below

The web build went to a temporary directory instead of the root dist/, because a running sidecar serves that directory.

In the full root run packages/hive-mind-cli/tests/cli-help.test.ts also failed once: the isolated home made npm print an update notice into the CLI output. The setup now turns that notice off and the file passes (3 of 3), so the totals above count it as passing.

Known failures (macOS)

All 41 failing root tests are in four files and fail the same way when run alone. They are logged as UF-023.

File Failing Symptom
packages/server/tests/local/managed-ollama-runtime.test.ts 38 of 45 "Official runtime archive did not contain ollama.exe"
packages/hive-mind-hooks-openclaw/tests/handler.test.ts 1 of 16 Expects /var/folders/..., receives /private/var/folders/...
packages/server/tests/local/files.test.ts 1 of 48 "escapes workspace root through symlink" in a move test
packages/server/tests/local/external-tool-runs.test.ts 1 of 17 "OpenClaw run did not settle" (timeout, not investigated)

Checkpoint: end of week 1

Measured on 2026-10-20 at 5f08869, after P0.1–P0.3, P1.4, P1.6 and P2.2 part 1. Fewer tests than at the baseline is expected: the removed code took its tests with it.

Check Result
Package build, web typecheck, check:deps Pass
Lint, root 0 errors, 0 warnings
Lint, web 38 errors, 60 warnings (all pre-existing; down from 49 errors)
Unit tests, root 638 files, 10,751 tests: 10,655 passed, 41 failed, 55 skipped
Unit tests, web 212 files, 1,884 tests, all passed
E2E 410 passed, 18 skipped, 0 failed (7.0 min, Chromium)

The 41 failures are the same four files as in "Known failures" above.

E2E ran with WAGGLE_E2E_PORT=3344 WAGGLE_E2E_BASE_URL=http://127.0.0.1:3344 WAGGLE_E2E_REUSE_EXISTING_SERVER=0 WAGGLE_E2E_DATA_DIR=<temp dir> npm run test:e2e, so it neither reused nor disturbed a sidecar on :3333 (it still rebuilds the root dist/). The first run failed 19 tests, all on removed features: /api/tier, Stripe, /api/compliance, the billing settings tab, the payment pages and the upgrade dialog. Those specs now assert that the endpoints and pages are gone (404) instead. The 18 skips depend on the environment: Windows-only launcher lanes, live chat without a provider key, and the optional team server.

Checkpoint: end of week 2

Measured on 2026-10-24 on oss/community-edition, after P1.1, P1.2, P1.3b (macOS part), P1.5, P1.7 and P2.1. Memory runs on mem0 through the memory contract, and the excluded and postponed packages are neither built nor tested, so the root run has fewer tests again.

Check Result
Package build (npm run build:packages, now tsc -b), web typecheck, app typecheck, check:deps Pass
Test typecheck (npm run typecheck:tests, new in P2.1) 298 errors, all pre-existing; advisory in CI
Lint, root 0 errors, 0 warnings
Lint, web 36 errors, 59 warnings (all pre-existing)
Unit tests, root 441 files, 8,350 tests: 8,261 passed, 40 failed, 49 skipped (2.1 min)
Package-install tests (CLI, marketplace, launcher; run serially) 31 of 31 passed
Unit tests, web 213 files, 1,906 tests, all passed
E2E 405 passed, 11 skipped, 0 failed (6.4 min, Chromium)
Sidecar bundle Builds from source (7.0 MB); no Stripe, Clerk, BullMQ, ioredis, postgres, Drizzle, Baileys, hive-mind or sqlite-vec; check-sidecar-resources passes
Packaged sidecar smoke run Starts in 3 s on a temporary data dir, saves and finds a memory, creates no personal.mind

The 40 failures are three of the four known files: managed-ollama-runtime (38), files (1) and external-tool-runs (1). The fourth, the hive-mind OpenClaw hook handler, left the root run with the excluded packages.

The package-install tests run apart from the root run, as in CI: each one packs and installs packages with npm, and under load they time out. The e2e run used the same isolated port and data directory as in week 1. Two e2e journeys, B1.6 (first memory search under 500 ms) and B5.6 (oversized search query), failed once each when only their spec files ran; both passed in the full runs.

After updating Rust to 1.99.1, the Tauri shell builds for macOS arm64 (Waggle.app and Waggle_0.2.0_aarch64.dmg, unsigned), and its 30 Rust tests pass. The first packaged launch stayed on "Starting Waggle…" because the shell looked for the sidecar next to the executable (UF-004); with that fixed, the app starts its bundled sidecar on a temporary data directory and opens the onboarding screen.

Checkpoint: macOS acceptance, week 3

Measured on 2026-11-01 on oss/community-edition (from 4d54612 to a70bc14), macOS arm64, after P1.3c, P1.3d, P2.2–P2.5 and the version bump to 1.0.0.

Check Result
Root unit tests The 40 known failures (managed-ollama-runtime 38, files 1, external-tool-runs 1). Under load git-tools, background-bash, cli-runtime and skill-watcher can fail on timing; alone they pass
apps/web unit tests 1,914 passed
Rust (cargo test --lib) 33 passed
E2E 400 passed, 8 skipped, 5 failed. All 5 were phone-width tests that waited on the sidebar, which phones no longer show (UF-057); after fixing the tests they pass (7 of 7 in the rerun)
Desktop app Waggle.app and Waggle_1.0.0_aarch64.dmg build; scripts/release-smoke.mjs desktop starts the app and its sidecar answers /health
Server installer Installed from the packed archive: doctor passes, run serves the UI, update and uninstall work, the data directory stays

Acceptance against section 5 of the scope, as a new user of the server installer on a fresh data directory (the managed Ollama runtime and Llama 3.1 8B copied in, so nothing was downloaded):

Scope item Result
Agent runtime with several personas Researcher, Writer, Analyst, Planner and others; the onboarding template opens the workspace with its persona
Web UI first, CLI as a headless alternative Web UI: onboarding, chat, memory, agents, Skills Hub, marketplace and settings work. CLI: answers with a local model after the fix below
Skills, marketplace, connectors Marketplace lists skills, connectors (Gmail, Google Drive, Notion and others) and MCP servers; starter skills and packs appear after UF-066
Local model out of the box, own cloud keys Onboarding finds the local model; chat answers with Llama 3.1 8B; 12 providers take a key in Settings → Models
Persistent memory The agent saved a fact through save_memory; after a server restart a new question recalled it ("Rust, according to prior information stored about you")
No billing, tiers or upsell Settings has General, Models, Permissions, Channels, Backup and Advanced; no plan, team or enterprise tabs
Analytics only with consent A source build has no PostHog key: no consent prompt and no request to PostHog

Found and fixed during acceptance:

  • Installed builds had no capability packs and no starter skills (UF-066).
  • The status bar named an unusable default model after a restart (UF-067).
  • The CLI could only reach models through the LiteLLM proxy, which the community edition no longer starts; local ollama/* models now go to Ollama directly. Cloud models in the CLI still need a LiteLLM-compatible proxy.

Noted, not fixed:

  • The first local-model turn took about 2.5 minutes: Llama 3.1 8B loads into memory, and the first request carried about 16,500 input tokens. Later turns were much faster.
  • Memory is mem0 (decision Q1). Whether GluoMem becomes the default is open with the owner.

Not run yet

  • Visual regression (npm run test:visual). It compares screenshots against 36 baselines with a 0.3% pixel threshold; the platform they were captured on is not recorded, and the Cockpit and Memory screens have changed since. It is re-baselined in P2.7.
  • Infrastructure suites (npm run test:infra) need Postgres and Redis and cover team mode, which is not part of the community edition.
  • Windows and Linux. This record covers macOS only.