ADHD v2 — Milestone 1: core + lint#10
Open
hhff wants to merge 35 commits into
Open
Conversation
…tecture Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ths, manifest in M5 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…pping Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…eview fix) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Implements core/tokens.ts with Domain, Mode, Token, StyleShell, Snapshot types and domainOf() function; core/snapshot.ts with canonicalize() and contentHash() for deterministic snapshot hashing. All tests pass (3/3). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
….dark matcher Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… wording (review fix) - Fix 1: resolveCssEntry anchors cfg.cssEntry to dir with join() and verifies existence with existsSync() - Fix 2: Validator accepts cssEntry: null explicitly (was throwing before) - Fix 3: Improved error messaging for fileKeyFromUrl and replaced readFileSync probes with existsSync Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…runtime Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… (review fix)
Rogue-collection variables now get path "rogue:<collection>/<name>" instead of
their plain name, so a variable in a non-Primitives/Semantic collection can
never key-collide with a real primitives token under the ${collection}/${path}
matching convention (snapshot.ts, css.ts). Real token paths are lowercase
alnum/hyphen/slash only and can never contain ":", so this makes collision
impossible.
Also adds minimal inline-doc tests to extract.test.ts covering: corrupted
final-chunk hash, dangling alias, unknown Semantic mode name, and
case-insensitive collection/mode name matching — without touching the large
committed recording fixture.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…a fixtures
Faithful TS port of lib/lint-engine/structure-checker.js's visit() walker,
push() helper, deepLink(), AUTO_NAME_RE, SINGLE_CHILD_SHAPE_EXEMPT, and
isVisiblePaint(), carrying every hard-won exemption (STRUCT001 single-shape
exemption; STRUCT002/003/005 skip COMPONENT_SET; invisible paints/effects
excluded; empty boundVariables{} isn't a binding; STRUCT004 accepts style OR
fontSize; auto-derived variant names exempt from STRUCT009).
Deliberate change: STRUCT009 now uses core/naming's strict caseMatches
(no more loose /^[a-z0-9-/.]+$/ fallback). Path-like identifiers (variant
property names, e.g. "avatar/size") are split on '/' and '=' and each
segment case-checked independently via a new caseMatchesPath helper.
Test: transcribed v1's struct-fixtures.test.js fixture->rule table verbatim
into test/struct.test.ts against the copied real-Figma fixtures; all 11
cases pass unchanged (verified against v1's checker directly as ground
truth — no expectation had to move for the strictness change).
…suites
Removing "type": "module" from plugins/adhd/package.json fixes a cross-cutting
regression where Node treats v1's CommonJS files under plugins/adhd/lib/** as
ESM, causing the v1 test suites to fail. Nothing in v2 requires this field:
build.mjs is explicitly ESM via its .mjs extension, vitest handles TS/ESM itself,
and dist/ has its own {"type":"commonjs"} marker.
Verified: v1 tests 283/283 pass (96+103+67+17), v2 tests 53/53 pass, typecheck 0,
build fresh.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ral; pin rename 1:1 guarantees (review fix)
…full kind coverage (review fix)
- Standalone hex now only counts inside string-literal spans on the line
(computed via a line-local scanner), never in comments/JSX braces/URLs.
- tokensByColor/tokensByDimension Maps were built but never consulted;
replaced with real precomputed { path, rgba } / { path, px } arrays that
findNearestToken compares against directly (no re-parsing per match).
- Collapsed duplicated exact->near lookups into matchColorToken + rgbaClose,
and removed the dead color->dimension fallthrough (color parse now
returns early).
- Added test coverage for inline-hex, mixed arbitrary-class + inline-hex
lines, near-match epsilon, and hex-outside-any-string (no finding).
…, pretest build (review fix)
- readChunks wraps readdirSync/readFileSync/JSON.parse and rethrows AdhdError
naming the failing path + cause, with a "check --figma-chunks points at a
directory of valid chunk JSON files" fixup; an empty dir gets the same
treatment ("no chunk files found in <dir>").
- runLint catches assembleChunks' fixup-less AdhdErrors (e.g. the contiguity
check on unpadded chunk filenames) and fills in a zero-padded-naming fixup
so the CLI always prints a → hint.
- Off-system file reads (git ls-files result) are now defended per-file: an
unreadable tracked file is skipped with a stderr note instead of crashing
the whole lint run.
- package.json: added "pretest": "npm run build" so vitest's subprocess tests
always run against a freshly built dist/.
…eview fix) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…xup, dead import, $WORK cleanup
Important: - diff/css: compare bare semantic color tokens (--background etc., domain 'other') as colors, not exact strings — kills false value-drift on oklch↔hex - payload: flag off-grammar Figma variable names unsyncable (was CI-breaking existence drift) - payload: flag Primitives collections with >1 mode unsyncable (was silent loss) - cli: route non-AdhdError failures + --out write errors through exit-2 contract - off-system test: actually exercise the adhd:off-system skip Minor: - payload: flag unresolved bound-primitive ids on the style shell - cli: drop dead 'export' on fail(); remove 3 unused test imports - report test: assert no timestamp instead of tautological double-call - example: add adhd.config.json so /adhd:lint works from the demo consumer Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Important: - css: @theme inline (Layer-3 exposure vars) no longer modeled as primitives — kills ~7 false existence errors on the example's own globals.css - diff: tighten probable-rename pairing (equal length, one differing segment) — stops false-pairing same-valued primitives like color/black vs color/ink - off-system: normalizeDimension parses rem (x16) — nearest-token suggestions now fire for Tailwind's rem-based scale - lint/report: surface StyleShell.unsyncable reasons into the Cannot-sync section (were computed but silently dropped); note git-unavailable off-system skips - struct: guard node.name before split() — a nameless node no longer crashes the whole lint run - diff test: exercise colorsEqual with a real oklch value (was hollow a===b) Minor: duplicate-collection-name map collision -> cannotSync; describeValue renders mixed alias/literal modes; struct severity assertion; dead exports dropped; bindingless catch in lock.ts; unreachable ternary in payload. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Important: - diff: unify typography dimensions (rem↔px FLOAT) like spacing/radius — idiomatic Tailwind text-size/leading tokens no longer false-drift - css: guard normalizeColor against bare numbers — culori parses '500' as #550000; font-weight/unitless-spacing tokens were being corrupted to hex - diff: a lock-driven definite rename now still reports coincident value/ structural drift instead of silently swallowing it Minor: - lock: validate nested lock shape, throw the friendly 'Delete adhd.lock.json and re-sync' error instead of a downstream cryptic TypeError Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Important: - diff: make dimension-equivalence a UNIVERSAL fallback instead of a domain whitelist — border-width/size/blur/etc. dimension tokens no longer false-drift (px<->rem<->unitless unified for any domain; retires the recurring theme) Minor: - css: classify() detects @theme inline by token membership, so combined modifiers (@theme inline static/reference) are correctly treated as exposure - lock/config: a present-but-unreadable file (EACCES/EISDIR) now raises a real IO error instead of being misreported as 'file not found' Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Minor (robustness): a chunk file containing bare `null` is valid JSON but not a chunk — readChunks now rejects non-object parsed values, and assembleChunks guards defensively, so it surfaces as a clean AdhdError with the chunk-dir fixup instead of a raw 'Cannot read properties of null' TypeError. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Member
Author
Adversarial review campaign — converged ✅Ran a 6-round adversarial review loop: each round fanned out 6 skeptical finders (dead-code, correctness, integration, no-silent-loss, test-quality, robustness) over the whole branch, and every finding was independently verified by a second agent trying to refute it before it counted. Only reproducible defects (concrete input → wrong output) or provably-dead code survived.
39 verified defects fixed. Highest-impact (each would have made
Also fixed: off-grammar Figma names / extra Primitives modes / unresolved style bindings now surface in the Cannot-sync report instead of being silently dropped; exit-2 operational contract closed for unreadable/empty/misordered/ Final state: v2 140/140 tests, typecheck clean, committed Remaining merge gate (unchanged)Not exercised against a live Figma file (no MCP in the build env). Before merge, run |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What this is
Milestone 1 of the ground-up ADHD v2 rebuild (spec:
docs/superpowers/specs/2026-07-05-adhd-v2-rebuild-design.md, plan:docs/superpowers/plans/2026-07-05-adhd-v2-m1-core-and-lint.md).v2 replaces v1's LLM-as-plumbing skills and triplicated hand-rolled parsers with a single bundled TypeScript CLI. This milestone lands the core + a working
/adhd:lint.What's in it
dist/(freshness-gated in CI), vitest, strict TS.globals.cssreader (per-mode aliases, no silent loss),adhd.config.json+adhd.lock.jsonloaders.figmaruntime harness.adhd figma-script extract+adhd lintwith a strict exit-code contract (0 / 1-check / 2-operational+fixup)./adhd:lintrewritten as a 44-line thin wrapper.Verification
Known merge gate
Not yet exercised against a live Figma file (no MCP in the build env): confirm
use_figmareturns theExtractChunkshape the CLI expects and thatserializeStyle's field names match the real Figma styles API. Run/adhd:lintonce against the real configured file before merging.🤖 Generated with Claude Code