feat: Wave 3 v1.0.0 rollout — VS Code ext, Claude Code skill, Mistral+Cohere, --latency, playground showcase

[faraa2m]

Summary

Wave 3 of the v1.0.0 rollout. Five workstreams, all additive — no breaking changes.

What's in here

VS Code / Cursor extension (packages/vscode/, new package)

• Status bar shows live <model> · <tokens> tok · $<cost> for the active editor file.
• Settings: tokenometer.model, tokenometer.format, tokenometer.warnOnCostAbove (turns the bar warning-yellow when exceeded).
• Commands: Tokenometer: switch model (quick-pick across KNOWN_MODELS), Tokenometer: show info.
• Debounced 200ms recompute on doc change; 1MB byte-size gate; supported file types .md/.txt/.json/.yaml/.yml/.xml.
• Reuses @tokenometer/core via esbuild bundle. Marketplace publish lands in Wave 4 Phase I.
• 22 unit tests for the pure helpers (status bar formatter, threshold check, debounce, file gate).

Claude Code skill (packages/claude-code-skill/, new package)

• SKILL.md with name: tokenometer-cost-check + multi-trigger description.
• 13-section body teaches Claude Code agents when + how to invoke npx tokenometer: offline + empirical mode, --output json/sarif, --by-file, --image, auto provider detection, .tokenometer.yml, tokenometer init for CI scaffold.
• One-line install: ~/.claude/skills/tokenometer/SKILL.md. Convenience install.sh included.

Mistral + Cohere providers (@tokenometer/core)

• Mistral: mistral-tokenizer-js for SentencePiece family (Mistral 7B, Mixtral 8x7B/8x22B, Mistral Large 2407, Codestral 22B); chars/4 heuristic for Tekken-tokenizer models (NeMo, Pixtral, Mistral Small 2409+, Devstral, Mistral Medium 2505+, Magistral, Ministral). All marked approximate: true. Empirical mode unsupported — Mistral has no public token-count API; throws a clear error pointing at the metered chat-completion fallback.
• Cohere: chars/4 heuristic offline. Empirical via POST /v1/tokenize using bare fetch (Cohere SDK ships only a REST wrapper, no offline tokenizer — saved 4.9 MB install).
• Pricing: Mistral auto-sourced from @tokenlens/models/mistral (19 models). Cohere via LOCAL_OVERRIDES (command-r, command-r-plus) until tokenlens adds upstream.
• KNOWN_MODELS count: 42 → 63.

--latency flag (CLI + core)

• Measures real generation latency: TTFT + total ms + tokens/sec. Reports p50 / p95 / mean over n trials (default 3, configurable 1-10 via --latency-trials).
• Implies --empirical. Bumps default --max-spend ceiling to $0.25 to cover ~200-token generations.
• Streamed via SDK for Anthropic + Google; bare fetch SSE/NDJSON for OpenAI + Mistral + Cohere (avoids extra SDK install footprint).
• Table renderer adds latency columns when present. JSON + SARIF outputs carry the full latency block.
• Differentiation pitch: the only LLM cost CLI that also reports latency. No competitor (tokencost, tiktoken, gpt-token-counter-live) ships latency.

Playground showcase (packages/web/)

• 11 new routes via react-router-dom@7: /diff, /by-file, /sarif, /vision, /config-builder, /init, /models (Cost Atlas: sortable + searchable + filter-by-provider), /models/<id> (per-model SEO pages), /editor, /claude-code, plus / and /calculator for the existing landing.
• Top nav with Tools dropdown; footer linking repo + npm + Marketplace placeholders.
• Map<Provider, ...> everywhere instead of static Records — playground gracefully extends as more providers land.
• Static robots.txt + sitemap.xml; per-page document.title via usePageTitle hook (no react-helmet dependency).

Why now

• Plan called for Wave 3 to add the user-facing differentiation: editor presence (E.1), agent-AI awareness (E.2), provider breadth (H), latency (the marketing line), and a playground that demos every shipped feature.
• All five landed in this PR. v1.0.0 launch can claim every feature it advertises.

Test plan

• npm run lint — 89 files clean
• npm run typecheck — clean
• npm test — 227/227 tests across 20 files (was 150/16 before Wave 3)
• npm run build — clean
• npm run build -w @tokenometer/web — clean (4.43 MB main JS, 2.01 MB gzipped, larger-than-500kB warning is acknowledged + accepted for now)
• CLI smoke: SARIF output validates as version: "2.1.0"; --by-file / --output json / --image paths exercised by integration tests
• Action smoke (post-merge): tokenometer's own prompt-cost.yml runs and the per-file Action comment renders
• Manual: install the VS Code extension .vsix (built locally via npm run package:vsix --workspace=packages/vscode) and confirm status bar renders. Marketplace publish deferred to Phase I.
• Manual: cp packages/claude-code-skill/SKILL.md ~/.claude/skills/tokenometer/ then issue a test prompt to confirm Claude Code triggers on it

Out of scope (Wave 4)

• Phase I unified release pipeline — VS Code Marketplace publish via vsce, Open VSX publish, post-publish smoke test job, Vercel deploy hook, Marketplace verify HTTP probe. Right now the release workflow only handles npm via Changesets.
• Phase G content posts — drafts are pre-staged in ~/Downloads/tokenometer-content-todo.md; manual posting after Wave 4 is green.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
tokenometer	Ready	Preview, Comment	May 10, 2026 4:10am