Consolidate salvaged branches: signal_mesh + MCP↔WS bridge + pomodoro demo, green CI with per-package coverage ratchet

[MelbourneDeveloper]

TLDR

Consolidates all useful work salvaged from the repo's stray branches onto one branch — two new packages (signal_mesh, mcp-websocket-bridge), a pomodoro demo across the examples, CI/tooling hardening — and gets make ci fully green with per-package coverage that auto-ratchets up.

What Was Added?

• signal_mesh/ — new Dart package: peer-to-peer encrypted mesh messenger. Kademlia DHT peer discovery, X3DH + Double Ratchet end-to-end crypto, phone-number identity + attestation, store-and-forward queue, and a transport abstraction. Ships its own test suite (crypto, node_id, kademlia, routing_table, identity, message, store_forward, transport). publish_to: none (path deps on dart_node_core/dart_node_ws).
• packages/mcp-websocket-bridge/ — new TypeScript package bridging MCP over WebSocket (bridge/transport/session/http-client/types/errors), with SPEC.md, README, and an examples/chat-bridge.ts.
• Pomodoro demo across examples/ — backend routes + PomodoroService + validation schema, mobile pomodoro_screen, shared PomodoroSession model / theme styles / HTTP client, plus backend and mobile tests. Also a paper_demo_screen (react-native-paper).
• Agent/dev tooling — .claude/skills/* (build, ci-prep, code-dedup, fix-bug, fmt, lint, spec-check, submit-pr, upgrade-packages, website-audit), assistant rule files (.cursorrules, .clinerules, .windsurfrules, copilot-instructions), .deslop.toml, .editorconfig.
• coverage-thresholds.json — single source of truth for per-package coverage floors that ratchet UP only.

What Was Changed or Deleted?

• Coverage now actually enforced. coverage-thresholds.json was a single default_threshold: 70. It now carries a per-package map seeded from the legacy GitHub Actions variable MIN_COVERAGE=70 and auto-ratcheted in tools/test.sh: after every fully-green run each package's floor is raised to its measured coverage and never lowered (default_threshold raised to a 90% library floor for new packages).
• make ci made green (these were real pre-existing failures, not introduced here):
  • packages/dart_node_coverage/lib/src/runtime.dart — removed all illegal as casts and a bare catch; replaced with typed @JS extension types (JSON.parse, Object.keys) and generic getProperty<T>.
  • examples/backend/server.dart — the pomodoro routes were a broken merge (missing models/pomodoro.dart import; calling an undefined getAuthContext; unmatched Results). Rewrote all 10 routes to the same getAuthContextWithService + getValidatedBody pattern the task routes use.
  • dart_node_mcp / dart_node_vsix / dart_node_ws test files — cascades, the JSArray prefer_is_empty false-positive (.toDart.isNotEmpty), unnecessary_async, a tearoff, nullable-cast cleanup.
  • cspell-dictionary.txt — added the legitimate technical terms cspell flagged (Kademlia, HKDF/HMAC, ciphertext, otpk, prekey, pomodoro, deslop, worktree, automemory).
• Removed tools/build/ (build.dart, add_preamble.dart) — superseded.
• analysis_options.yaml tightened (strict-casts/strict-inference/strict-raw-types); Makefile standardized (lint/test/fmt/ci); .github/workflows/ updated (CI reads the coverage threshold from the JSON; deploy-website → deploy-pages); website gained pub.dev + GitHub links and a getting-started page.

How Do The Automated Tests Prove It Works?

• make ci is green end-to-end (MAKE_CI_EXIT=0): cspell reports 0 issues in 0 files; dart analyze reports 0 errors across every package and example; all test tiers pass with coverage enforced — dart_node_core/dart_node_ws/dart_node_express/dart_node_better_sqlite3/dart_node_mcp/dart_node_react_native/dart_logging at 100%, web_counter 98.5%, reflux 97.6%, markdown_editor 94.8%; dart_node_react/frontend browser tests pass under Chrome.
• Ratchet verified across two runs: run Add Web Sockets And Fix RN App #1 logged ⬆️ Ratcheted … → 100.0% (was 70%) for every package and Coverage thresholds raised; run Adds Website with Documentation and Examples #2 produced no ratchet lines and still passed — proving the floor is monotonic and idempotent. A subsequent coverage drop below the stored floor now fails make test.
• The runtime.dart refactor is exercised at runtime, not just statically: the Node coverage CLI uses writeCoverageFile/_mergeData to merge coverage for every NODE_PACKAGES run, all of which passed.

Spec / Doc Changes

• CLAUDE.md — TDD rule + deslop / branch-discipline / autonomy / auto-memory sections.
• coverage-thresholds.json _doc rewritten to describe the auto-ratchet contract.
• New signal_mesh/README.md and mcp-websocket-bridge/SPEC.md + README.

Breaking Changes

• None to any published package surface. The two new packages are not published (publish_to: none / path deps); all other changes are examples, CI/tooling, and lint-clean refactors with no behavioral change.

Note: signal_mesh and mcp-websocket-bridge ship with their own tests but are not yet wired into the tools/test.sh tier runner, so they are not gated by make ci in this PR — a sensible follow-up.