v0.2.1 delight pass: color hierarchy, warmer empty state, home-relative paths

[AnobleSCM]

Summary

Strictly-scoped presentation polish on the local stack report, releasing as v0.2.1. Exactly three things, per spec:

1. Terminal color hierarchy (TTY only). Type labels (mcp/plugin/skill/subagent) get a single cyan accent; per-type counts are bold. Header/footer unchanged from today.
2. Empty state. Warmer, honest headline (No AI tooling detected on this machine yet. — reused verbatim from the existing --markdown empty-state sentence for consistency, rather than inventing new wording). Looked in: paths now print home-relative (~/.claude/skills); paths outside $HOME stay absolute.
3. Release prep. 0.2.0 → 0.2.1 in package.json and CLI_VERSION, plus a CHANGELOG entry.

Do not merge — draft only, per instructions. Not published to npm.

Taste call: one shared accent, not one per type

Went with a single cyan accent for all four type labels rather than a distinct hue per type. Reasoning:

• Four hues across up to 4 rows per client block starts to read as a legend, not a restrained CLI (the explicit "not rainbow soup" target).
• cyan is already the established accent in this codebase (used for the verification URI in the sync device-flow prompt) — extending it keeps one stable accent-to-meaning mapping instead of inventing a second convention.
• The type words are already text-distinguished from each other; color's job here is separating "label" from "count" and "name list," not distinguishing the four types from one another.
• Matches restrained modern-CLI conventions (pnpm/vite/biome: one accent + bold + dim, not a hue-per-category legend).

Invariant evidence

All diffs below are v0.2.0 (origin/main, built fresh) vs this branch, run against byte-identical fixture input (same fixture $HOME + cwd, both dist builds invoked directly).

(a) Piped (non-TTY) main-report text — byte-identical except the empty-state copy change

$ diff old.default.txt new.default.txt   # rich case, 7-tool fixture
IDENTICAL

$ diff old.empty.txt new.empty.txt       # empty case — the deliberate exception
1c1
< ✓ No AI tooling detected.
---
> ✓ No AI tooling detected on this machine yet.
9,16c9,16
<   - /…/fixture-home-empty/.claude.json          (8 lines, all absolute)
---
>   - ~/.claude.json                                (same 8 lines, home-relative)

Also verified on this machine's real $HOME (~121 real tools) — piped rich-case output is byte-identical, confirmed with diff (see "Sanity runs" below).

(b) --json / --markdown — byte-identical except the cli_version field

$ diff old.json.txt new.json.txt
2c2
<   "cli_version": "0.2.0",
---
>   "cli_version": "0.2.1",

$ diff old.markdown.txt new.markdown.txt
IDENTICAL   # markdown carries no version field, so no exception at all here

$ diff old.empty.json.txt new.empty.json.txt   # empty case too
2c2
<   "cli_version": "0.2.0",
---
>   "cli_version": "0.2.1"
# paths_checked stays absolute — the display-only home-relativization never
# touches --json, confirmed by a dedicated test.

stderr streams (truncation warnings) are also identical across all modes — confirmed empty in this fixture (no truncation).

(c) README demo/example verbatim-match invariant
assets/devcat-report.svg is untouched (git diff origin/main -- assets/devcat-report.svg is empty). I manually line-matched its <tspan> text content against the README's "same output as text" block — still byte-identical, since color is TTY-only ANSI and never touches text content (this is now also a standing test: "colorizing never changes the underlying text — stripped output matches the plain render").

One thing to flag: the SVG's fill colors (#c9d1d9 flat gray body text) reflect the pre-delight-pass monochrome scheme. A real colorized TTY run now differs visually from the static demo (cyan labels, bold counts). Regenerating the SVG is conditioned in the brief on visible text changing, which it hasn't — so I left the asset alone rather than reverse-engineer its original generation pipeline. Flagging so it's a deliberate call, not an oversight, in case a follow-up wants to regenerate it.

(d) README sentences
Only one edit: the illustrative "cli_version": "0.2.0" literal inside the --json example block → "0.2.1", since real --json output will say 0.2.1 the moment this ships. No other README sentence references anything this pass touched (grepped for the old empty-state copy — it isn't quoted anywhere in the README).

Explicitly not touched, flagged rather than fixed: the README's "Profile sync — paused while devcat.dev is rebuilt" framing is drifting stale now that the static landing page is live (per the knowledge-plane rider expecting this to "ride along with 0.2.1"), but rewording it touches the profile-sync/devcat.dev framing the freeze record governs and isn't one of the three named items — left for a separate, deliberate pass rather than folded in here.

(e) No new network paths; skills/subagents-never-in-sync-payload guarantee untouched
Nothing under src/api/, src/auth/, src/commands/sync.ts, or src/manifest/index.ts (home of SyncableToolEntry/syncableTools) appears in the diff — confirmed via git diff origin/main --stat. Full file list touched: CHANGELOG.md, README.md, package.json, src/lib/isHeadless.ts, src/ui/report.ts, src/version.ts, plus 6 test files.

Gates (all green)

npm test    → 246/246 passed (238 pre-existing + 8 new/updated)
npm run lint → clean
npm run build → clean
npm pack --dry-run → 67 files, identical file LIST to a fresh v0.2.0 pack (diffed)

New tests specifically added for this pass: TTY-color vs non-TTY plain (report.tty-color.test.ts), NO_COLOR gating including the empty-string edge case (colors.test.ts, isHeadless.test.ts), and empty-state copy + home-relative path display (report.test.ts).

Sanity runs on this machine

Rich case (real $HOME, 121 tools) — before/after diff is byte-identical:

✓ Your AI-coding stack — 121 tools

Claude Code · 111 tools
    4 mcp      atelier-board, codegraph, mcphub-gateway, workshop
   44 plugin   accessibility-compliance, agent-orchestration, clangd-lsp, …
   56 skill    ask-fable, ask-matt, clarify, codebase-design, …
    7 subagent clean-code-reviewer, code-reviewer, debugger, …

Codex · 9 tools
    3 mcp      computer-use, node_repl, openaiDeveloperDocs
    6 skill    claude-workflow-compat, design-taste-frontend, humanizer, …

Cursor · 1 tool
    1 mcp      posthog

121 tools in Claude Code, Codex, and Cursor · 13 locations checked

(identical before and after — piped output is unaffected by the color change, as required)

Real colorized TTY output, captured via script(1) against a real pty (cat -v rendering so the escape codes are visible as text):

^[[1mClaude Code^[[22m ^[[2m· 111 tools^[[22m
  ^[[1m  4^[[22m ^[[36mmcp      ^[[39matelier-board, codegraph, mcphub-gateway, workshop

— bold count (^[[1m...^[[22m), cyan type label (^[[36m...^[[39m), plain name list. Matches the design exactly.

Empty case (HOME → empty temp dir) — before/after:

BEFORE: ✓ No AI tooling detected.
AFTER:  ✓ No AI tooling detected on this machine yet.

BEFORE:   - /Users/andrewnoble/.claude.json
AFTER:    - ~/.claude.json
(cwd-relative project paths outside $HOME — e.g. …/real-cwd/.mcp.json — stay absolute in both)

Judgment calls

1. One shared cyan accent, not one hue per type — see "Taste call" above.
2. Reused the existing --markdown empty-state sentence verbatim for the terminal empty state, rather than inventing new copy — it was already the warmer, "yet"-qualified phrasing the original author chose for that sibling code path; consistency over a third distinct phrasing.
3. Fixed NO_COLOR='' (empty value) not counting as "set." !!process.env.NO_COLOR treated an empty-string value as unset, contradicting the no-color.org standard this README already cites ("disable color when [NO_COLOR is] present, regardless of its value"). Judged in-scope because the brief explicitly pre-authorizes NO_COLOR correctness work ("presentation-correctness and in scope") and the fix sits directly in the TTY-detection module the brief names. Updated the one existing test that had locked in the old (incorrect) behavior.
4. Did not touch the stale "paused while devcat.dev is rebuilt" copy — see invariant (d) above. Flagged, not fixed.
5. Did not regenerate assets/devcat-report.svg — see invariant (c) above. Flagged, not fixed.
6. Restored an incidental package-lock.json diff (npm-client peer-metadata normalization noise from npm install, zero actual dependency changes) rather than committing it — kept the diff surgical.
7. src/ui/render.ts (sync/device-flow renderers: renderSuccessSummary, renderEmptyManifest, renderUserCodePrompt) was left untouched — the task's CONTEXT section describes only the default report command's current output, and render.ts serves the separate, currently-paused sync feature.

Housekeeping note (unrelated, no action taken)

Noticed a stale, already-merged worktree (.worktrees/polly/readme-truth-pass, commit 8357786, one commit behind the tip of the README-polish PR that's already on main) sitting in the local checkout this session started from. Didn't touch it — flagging only in case it's worth pruning.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.