Skip to content

v1.7.0 — state & resume, scope dial, preflight honesty, five new skills, hooks, corpus CI - #13

Merged
blyatiful1 merged 2 commits into
mainfrom
claude/plugin-analysis-improvements-37rbvz
Aug 14, 2026
Merged

v1.7.0 — state & resume, scope dial, preflight honesty, five new skills, hooks, corpus CI#13
blyatiful1 merged 2 commits into
mainfrom
claude/plugin-analysis-improvements-37rbvz

Conversation

@blyatiful1

Copy link
Copy Markdown
Owner

What this is

A 10-lens adversarial analysis of the plugin (7 Opus 5 judgment lenses — ADHD-user experience, first-time-user journey, token economics, failure modes, competitive position, corpus architecture, wild ideas — plus 3 Sonnet 5 mechanical sweeps) produced 89 findings. This PR implements the convergent ones, end to end. The plugin grows from 75 to 80 skills and gains its first hooks, scripts, and CI.

State & resume — the biggest convergent theme (4 lenses hit it independently)

  • New skill status owns design/PROGRESS.md: current phase, an always-accurate Waiting on you line, next-checkpoint ETA plus minutes-of-your-time, measured per-phase durations, and the still-cheap-to-change decisions. Status questions are answered from the file, never from transcript scrollback.
  • Root §Resuming an interrupted build: a reconstruction ladder (PROGRESS.md → artifact presence → open checkpoints → never re-run a completed phase). iterate hands partial records to it; retrofit now refuses directories that already contain design/DIRECTION.md.
  • Git checkpointing: the pipeline commits at every phase boundary; the dev server gets a single owner (the Lead) with PID/port recorded.
  • scaffold is re-entrant (step 0 detects a half-built tree and resumes at the first incomplete step) and now moves design/ into the project root and commits the record — fixing the orphaned-artifacts and subdirectory-init defects.
  • checkpoint gains a per-checkpoint Your time column (~2–15 min each), the post-CP2 session map, a mechanical definition of "unattended" (a session property, never a patience judgment), and warmer piecemeal-feedback handling.

Scope & cost

  • §Scope tiers (sketch / standard / flagship): a one-page landing site no longer pays the twelve-phase price; gates weaken only where a tier row says so, in writing.
  • §Context discipline: each design/* artifact is read once per context — aimed at the ~154K-token resident prefix identified as the largest cost line in the showcase receipt.
  • Progressive disclosure: 39 skills split into a decision core + references/{example,composes}.md (~140KB out of resident load). award-canon restructured into ARCHETYPE-MAP.md (~3KB, the only file Phase 2 loads), INVARIANTS.md (now read verbatim by design-judge instead of paraphrased), and the dossier bank.

Preflight & truth

  • Phase 0 preflight: verifies skills resolve, toolchain, and Playwright before the interview. A missing browser now means a declared-up-front UNVERIFIED gate verdict — defined in gate-visual, gate-responsive, gate-accessibility, and pixel-qa — instead of an undocumented hard block discovered five hours in (the README had promised degradation no skill implemented).
  • ship: env/secret scans moved to working-tree greps with a clean-tree precondition (git grep saw only tracked files, so both audits passed vacuously on a normal build); the production smoke test now requires /studio to 404.
  • Frontmatter trigger collisions resolved: consent↔analytics ("cookie banner"), seo↔sitemap ("sitemap"), routing↔hidden-craft ("custom 404"), and a trigger-first rewrite of gate-code's description.
  • Install docs: the ~/.claude/skills/ clone path silently loaded the orchestrator without its 79 specialists — replaced with a local-marketplace install, and Phase 0 aborts loudly on a half-install.

Five new skills

Skill What it adds
status The "where are we" surface + resume anchor
preview Non-production Vercel URLs at CP4/CP6 — review on your own phone, noindex enforced, production stays ship's
assets Client-material intake — logo/photos/copy/brand guide inventoried in design/ASSETS.md; downstream skills read before inventing
identity The missing brand-mark producer — wordmark, monogram (favicon/manifest), OG template, public/brand/ as single source
studio Dev-only /studio construction-site dashboard read off disk at zero token cost, hard-gated to 404 in production

Meta layer

  • hooks/ (first hooks in the plugin): a PostToolUse grep that blocks taste banned-list violations at write time (~30ms instead of a Phase 11 Opus fix round; scoped to ultraweb builds only, verified against fixtures) and an agent-activity JSONL log that feeds /studio.
  • Cross-build taste fingerprint (~/.claude/ultraweb/): mockup and checkpoint verdicts persist as a tiebreaker — with a mandatory anti-profile "heretic seat" every round and 10-build decay so it can't become a rut. Optional pairwise tournament mode for the mockup round.
  • CAST.md: canonical fact sheet for the eight-client cast, extracted (not invented) from the corpus; the four real divergences it found (Ledger & Lane's jurisdiction, Loop & Thread's archetype/palette/routes, Tidepool's label) are fixed in place.
  • scripts/lint-skills.mjs + GitHub Actions CI: frontmatter shape, reference resolution, count-claim consistency, roster/README completeness, agent model pins, references/ integrity — enforced on every push (currently 0 failures, 6 advisory warnings).
  • stack/versions.json + scripts/verify-stack.mjs: machine-readable pins with a 30-day hard expiry, ending STACK.md's internal version contradictions.
  • README repositioned: leads with the guided session, adds a "Where it sits" section against v0/Lovable/Bolt, frames cost per scope tier instead of one scary anecdote, and promotes retrofit as the zero-risk first taste.

Verification

  • node scripts/lint-skills.mjs0 failures across all 80 skills (6 advisory warnings, all documented)
  • bash -n / node --check on all hooks and scripts; the antislop hook verified against dirty and clean fixtures (blocks with exit 2 + named taste lines, passes clean writes)
  • Every ultraweb:* cross-reference resolves; version bumped to 1.7.0 in plugin.json + marketplace.json

The full 89-finding analysis report was delivered to the session owner separately.

🤖 Generated with Claude Code

https://claude.ai/code/session_01EZzdneJ23VxgFgQNXXEMhG


Generated by Claude Code

…ls, hooks, corpus CI

Born from a 10-lens adversarial analysis (ADHD-user, first-time-user, token
economics, failure modes, competitive position, corpus architecture, plus
mechanical sweeps) that produced 89 findings. This release implements the
convergent ones:

State & resume
- New skill `status`: design/PROGRESS.md — where are we, waiting-on-you,
  next-checkpoint ETA + minutes-of-your-time, per-phase durations
- Root §Resuming: a reconstruction ladder so interrupted builds resume from
  the artifact record instead of rebuilding paid-for phases
- Git commits at every phase boundary; dev server gets a single owner
- scaffold is re-entrant (step 0 resume check) and moves design/ into the
  project root with the record committed
- checkpoint: per-checkpoint "Your time" column, session map after CP2,
  mechanical definition of "unattended", gentler piecemeal-feedback handling

Scope & cost
- §Scope tiers (sketch / standard / flagship) — a landing page no longer
  pays the twelve-phase price
- §Context discipline: each design/* artifact read once per context
- Progressive disclosure: 39 skills split into decision core +
  references/{example,composes}.md (~140KB out of resident load);
  award-canon restructured into ARCHETYPE-MAP / INVARIANTS / CANON
  reference files, design-judge now reads INVARIANTS.md verbatim

Preflight & truth
- Phase 0 preflight: skills resolve, toolchain, Playwright — a missing
  browser now means an up-front declared UNVERIFIED verdict (defined in
  gate-visual/-responsive/-accessibility and pixel-qa), not a dead end at
  hour five
- ship: working-tree secret/env scans (git grep saw only tracked files),
  /studio must 404 in production
- design/ location contract; retrofit refuses ultraweb sites; iterate
  hands partial records to the resume ladder
- Frontmatter trigger collisions resolved (consent/analytics, seo/sitemap,
  routing/hidden-craft, gate-code rewrite)

New capability skills
- `preview`: non-production Vercel URLs at CP4/CP6 — review on your own
  phone, noindex, production stays ship's
- `assets`: client-material intake — build with the client's brand instead
  of inventing over it
- `identity`: the missing brand-mark producer — wordmark, monogram,
  OG template, design/IDENTITY.md + public/brand/
- `studio`: dev-only /studio construction-site dashboard, zero tokens
  per refresh, hard-gated out of production

Meta layer
- hooks/: write-time anti-slop enforcement (taste banned list as a
  PostToolUse grep) + agent-activity log feeding /studio
- Cross-build taste fingerprint (~/.claude/ultraweb/) with a mandatory
  heretic seat and 10-build decay; optional pairwise mockup tournament
- CAST.md: canonical fact sheet for the eight-client cast (four found
  divergences fixed in place)
- scripts/lint-skills.mjs + CI: corpus invariants enforced on every push
- stack/versions.json + scripts/verify-stack.mjs: machine-readable pins
  with a 30-day expiry
- README repositioned: leads with the guided session, honest per-tier
  cost framing, marketplace-only install (the skills-dir clone loaded the
  orchestrator without its specialists)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EZzdneJ23VxgFgQNXXEMhG
@coderabbitai

coderabbitai Bot commented Aug 14, 2026

Copy link
Copy Markdown

Important

Review skipped

Too many files!

This PR contains 153 files, which is 53 over the limit of 100.

To get a review, reduce the PR to 100 files or fewer by splitting it into smaller PRs or changing its base branch.

Upgrade to a paid plan to raise the limit.

This review couldn't start because sufficient usage credits or metered capacity aren't available. Add credits or update usage-based reviews in the billing tab, then retry.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: be723c45-13f4-40d8-8b88-95a596fa01c3

📥 Commits

Reviewing files that changed from the base of the PR and between 433520b and bc265b8.

📒 Files selected for processing (153)
  • .claude-plugin/marketplace.json
  • .claude-plugin/plugin.json
  • .github/workflows/lint.yml
  • CAST.md
  • README.md
  • ROSTER.md
  • SKILL.md
  • STACK.md
  • agents/design-judge.md
  • agents/pixel-qa.md
  • hooks/antislop.sh
  • hooks/hooks.json
  • hooks/studio-log.sh
  • scripts/lint-skills.mjs
  • scripts/verify-stack.mjs
  • skills/analytics/SKILL.md
  • skills/analytics/references/composes.md
  • skills/analytics/references/example.md
  • skills/animejs/SKILL.md
  • skills/animejs/references/composes.md
  • skills/animejs/references/example.md
  • skills/app-structure/SKILL.md
  • skills/app-structure/references/composes.md
  • skills/app-structure/references/example.md
  • skills/assets/SKILL.md
  • skills/award-canon/SKILL.md
  • skills/award-canon/references/ARCHETYPE-MAP.md
  • skills/award-canon/references/CANON.md
  • skills/award-canon/references/INVARIANTS.md
  • skills/brief/SKILL.md
  • skills/brief/references/composes.md
  • skills/brief/references/example.md
  • skills/cart/SKILL.md
  • skills/cart/references/composes.md
  • skills/cart/references/example.md
  • skills/checkpoint/SKILL.md
  • skills/checkpoint/references/composes.md
  • skills/checkpoint/references/example.md
  • skills/command-palette/SKILL.md
  • skills/command-palette/references/composes.md
  • skills/command-palette/references/example.md
  • skills/component-api/SKILL.md
  • skills/consent/SKILL.md
  • skills/consent/references/composes.md
  • skills/consent/references/example.md
  • skills/content-cms/SKILL.md
  • skills/content-cms/references/composes.md
  • skills/content-cms/references/example.md
  • skills/copywriting/SKILL.md
  • skills/direction/SKILL.md
  • skills/direction/references/composes.md
  • skills/direction/references/example.md
  • skills/email/SKILL.md
  • skills/email/references/composes.md
  • skills/email/references/example.md
  • skills/footer/SKILL.md
  • skills/forms/SKILL.md
  • skills/forms/references/composes.md
  • skills/forms/references/example.md
  • skills/gate-accessibility/SKILL.md
  • skills/gate-accessibility/references/composes.md
  • skills/gate-accessibility/references/example.md
  • skills/gate-antislop/SKILL.md
  • skills/gate-antislop/references/composes.md
  • skills/gate-antislop/references/example.md
  • skills/gate-code/SKILL.md
  • skills/gate-code/references/composes.md
  • skills/gate-code/references/example.md
  • skills/gate-content/SKILL.md
  • skills/gate-content/references/composes.md
  • skills/gate-content/references/example.md
  • skills/gate-performance/SKILL.md
  • skills/gate-performance/references/composes.md
  • skills/gate-performance/references/example.md
  • skills/gate-responsive/SKILL.md
  • skills/gate-visual/SKILL.md
  • skills/hero/SKILL.md
  • skills/hero/references/composes.md
  • skills/hero/references/example.md
  • skills/hidden-craft/SKILL.md
  • skills/i18n/SKILL.md
  • skills/i18n/references/composes.md
  • skills/i18n/references/example.md
  • skills/icons/SKILL.md
  • skills/identity/SKILL.md
  • skills/imagery/SKILL.md
  • skills/imagery/references/composes.md
  • skills/imagery/references/example.md
  • skills/iterate/SKILL.md
  • skills/marginalia/SKILL.md
  • skills/marginalia/references/composes.md
  • skills/marginalia/references/example.md
  • skills/micro-interactions/SKILL.md
  • skills/micro-interactions/references/composes.md
  • skills/micro-interactions/references/example.md
  • skills/mockup/SKILL.md
  • skills/motion-language/SKILL.md
  • skills/motion-language/references/composes.md
  • skills/motion-language/references/example.md
  • skills/overlays/SKILL.md
  • skills/overlays/references/composes.md
  • skills/overlays/references/example.md
  • skills/page-transitions/SKILL.md
  • skills/page-transitions/references/composes.md
  • skills/page-transitions/references/example.md
  • skills/payments/SKILL.md
  • skills/payments/references/composes.md
  • skills/payments/references/example.md
  • skills/preview/SKILL.md
  • skills/pricing/SKILL.md
  • skills/pricing/references/composes.md
  • skills/pricing/references/example.md
  • skills/product-detail/SKILL.md
  • skills/product-detail/references/composes.md
  • skills/product-detail/references/example.md
  • skills/retrofit/SKILL.md
  • skills/routing/SKILL.md
  • skills/routing/references/composes.md
  • skills/routing/references/example.md
  • skills/scaffold/SKILL.md
  • skills/scaffold/references/composes.md
  • skills/scaffold/references/example.md
  • skills/scroll-motion/SKILL.md
  • skills/scroll-motion/references/composes.md
  • skills/scroll-motion/references/example.md
  • skills/seo/SKILL.md
  • skills/seo/references/composes.md
  • skills/seo/references/example.md
  • skills/set-design/SKILL.md
  • skills/set-design/references/composes.md
  • skills/set-design/references/example.md
  • skills/ship/SKILL.md
  • skills/ship/references/composes.md
  • skills/ship/references/example.md
  • skills/showpiece/SKILL.md
  • skills/showpiece/references/composes.md
  • skills/showpiece/references/example.md
  • skills/sitemap/SKILL.md
  • skills/status/SKILL.md
  • skills/studio/SKILL.md
  • skills/theme-worlds/SKILL.md
  • skills/theme-worlds/references/composes.md
  • skills/theme-worlds/references/example.md
  • skills/tokens/SKILL.md
  • skills/tokens/references/composes.md
  • skills/tokens/references/example.md
  • skills/typography/SKILL.md
  • skills/typography/references/composes.md
  • skills/typography/references/example.md
  • skills/ui-states/SKILL.md
  • skills/ui-states/references/composes.md
  • skills/ui-states/references/example.md
  • stack/versions.json

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.


Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

Comment @coderabbitai help to get the list of available commands.

…ffold tree, vercel major pin in manifest

The authoring workflow's final CAST.md write landed after the v1.7.0 commit
and reverted the fixed-in-place annotation — restored. Plus two wiring nits
from the agents' handoff notes: scaffold's folder contract now names
components/brand/ (identity's Phase-3 target), and the Vercel CLI major pin
that ship and preview share moved into stack/versions.json.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EZzdneJ23VxgFgQNXXEMhG
@blyatiful1
blyatiful1 merged commit 196dc02 into main Aug 14, 2026
3 checks passed
@blyatiful1
blyatiful1 deleted the claude/plugin-analysis-improvements-37rbvz branch August 14, 2026 14:14
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants