glovrex-proof-card

# proof-card One card renderer, three CLIs. This package collapses three separate implementations of the same riso-editorial share card (font embedding, riso ink palette, dashed rows, hero number, segmented bars) into one engine, then puts `ccwrapped`, `ghwrapped`, and `tokenops` on top of it. Glovrex F56 Bundle 3 (Rank2, Receipt / Proof-Card Suite). See `manifest.json` for the Glovrex app contract. ## Structure ``` renderer/ @greymoth/proof-card-renderer, schema -> SVG/PNG (see renderer/README.md) apps/ccwrapped/ Claude Code usage -> card (npm i -g @greymoth/ccwrapped is the published original) apps/ghwrapped/ GitHub profile -> card (npm i -g @greymoth/ghwrapped is the published original) apps/tokenops/ Claude Code token cost -> card + advice (npm i -g @greymoth/tokenops is the published original) _source/ read-only clones of the source repos this was extracted from (not shipped) self-verify.mjs generates one real card per app, asserts the files exist on disk ``` `apps/*` here are a **separate local integration**, not the published packages. The originals are untouched. Package names are suffixed `-proofcard` and marked `"private": true` for exactly that reason. ## Usage-tracker role split Three tools measure similar things. Here's the difference: - **inkdex** (not in this bundle, Chrome extension): measures ChatGPT/Claude usage in the browser, client-side, nothing sent. - **ccwrapped** (this bundle): measures Claude *Code* usage from local JSONL transcripts (`~/.claude/projects`). CLI, not a browser tool. - **tokenops** (this bundle): also reads `~/.claude/projects`, but for cost breakdown and optimization advice (cache hygiene, boundary routing, output discipline), not a usage summary. Shares no code with ccwrapped beyond a vendored copy of the same JSONL-parsing engine (each repo vendors its own copy today, a real dedup opportunity this pass didn't address). ## Try it ```bash npm install # installs workspace deps (renderer + all 3 apps) node self-verify.mjs # generates real cards, asserts files exist (slow: full-lifetime scan) node apps/ccwrapped/bin/cli.mjs --week --wrapped out/cc node apps/ghwrapped/bin/cli.mjs <github-username> --wrapped out/gh node apps/tokenops/bin/tokenops.mjs card # writes apps/tokenops/_out/*.svg + *.png ``` Each `--wrapped` call now writes **both** `.svg` and `.png`. The three source CLIs only ever wrote `.svg` and told you to screenshot it in a browser. Real PNG output (via `@resvg/resvg-js`) is the actual "renderer 一本化" deliverable, not just deduplicated SVG-string code. ## What's real vs known gaps (honest, per QUALITY BAR #1) - All three integrated apps run against real local data (Claude Code transcripts, live GitHub API) in this workspace. See the final report for screenshots taken during verification. - `self-verify.mjs` exit 0 = 10/10 real-file checks pass. - **Gap**: `renderer.test.mjs` checks structure (regex/string matches) and PNG magic bytes, not visual layout. It will not catch a text-overlap regression. Several were caught only by generating a PNG and looking at it (see `CONTRIBUTING.md`). Any new layout should be eyeballed, not just unit-tested. - **Gap**: `renderer`'s `rows`/`hero` layouts leave a blank gap on short-content cards (documented in `renderer/README.md`). Cosmetic, not a correctness bug. - **Gap**: root `npm test` (workspace-aggregated) fails in this environment with a `'node' is not recognized` error from npm's cmd.exe spawn (a Volta/PATH interaction on this machine, not a code defect). Run `cd renderer && npm test` directly instead. That path is green and is what every verification claim in this report is based on. - **Not done**: `mcp-loadout`'s config-slot card (a fourth candidate layout, ranked-list-of-servers) was read as prior art but not folded in. The task scope was the three CLIs, not a fourth app.