cue-ai 0.9.0 → 0.9.2

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "cue-ai",
-  "version": "0.9.0",
+  "version": "0.9.2",
   "description": "cue — Agent Profile Manager for Claude Code & Codex. Pick a profile, launch with the right skills, MCPs, and plugins.",
   "type": "module",
   "bin": {
@@ -13,19 +13,28 @@
     "cue": "bun src/index.ts",
     "test": "bun test",
     "typecheck": "tsc --noEmit",
+    "lint": "biome lint src",
     "build": "bun build src/index.ts --no-bundle",
     "build:bundle": "bun build src/index.ts --target=node --outfile=dist/cue.js",
     "prepack": "bash scripts/prepack-check.sh && bun run build:bundle",
-    "postinstall": "echo '' && echo '  ✅ cue installed. Run: cue init' && echo '  Docs: https://github.com/opencue/claude-code-skills' && echo ''"
+    "postinstall": "echo '' && echo '  ✅ cue installed.' && echo '  Next: run `cue shell install`  (activates the claude/codex shim — required to load profiles)' && echo '        then `cue init`        (pin a profile to this directory)' && echo '  Docs: https://github.com/opencue/cuecards' && echo ''"
   },
   "keywords": [
     "claude-code",
     "codex",
     "agent",
     "profile",
+    "profile-manager",
     "skills",
+    "claude-skill",
+    "agent-skill",
     "mcp",
+    "mcp-server",
     "ai",
+    "ai-agents",
+    "ai-coding",
+    "anthropic",
+    "claude",
     "cli",
     "developer-tools"
   ],
@@ -33,11 +42,11 @@
   "license": "MIT",
   "repository": {
     "type": "git",
-    "url": "https://github.com/opencue/claude-code-skills.git"
+    "url": "https://github.com/opencue/cuecards.git"
   },
-  "homepage": "https://github.com/opencue/claude-code-skills",
+  "homepage": "https://github.com/opencue/cuecards",
   "bugs": {
-    "url": "https://github.com/opencue/claude-code-skills/issues"
+    "url": "https://github.com/opencue/cuecards/issues"
   },
   "files": [
     "bin/",
@@ -59,6 +68,7 @@
     "install.sh",
     "get.sh",
     "README.md",
+    "CHANGELOG.md",
     "LICENSE"
   ],
   "dependencies": {
@@ -68,6 +78,7 @@
     "yaml": "^2.6.0"
   },
   "devDependencies": {
+    "@biomejs/biome": "^2.4.16",
     "bun-types": "^1.3.14",
     "typescript": "^6.0.3"
   }

package/profiles/_types.ts

@@ -50,6 +50,14 @@ export interface Profile {
   // composes via foldComposite. Recommendations are NOT inherited and do NOT
   // auto-merge skills/MCPs — purely a discovery hint.
   recommends?: string[];
+  /**
+   * Companion profiles that start CHECKED in the picker's combine-multiselect
+   * whenever this profile is the primary, regardless of cwd detection. Stronger
+   * than `recommends` (which only tags a row): use it when a profile is only
+   * useful alongside a specific companion (e.g. designer-medusa-vite always
+   * wants the medusa-dev backend). Still opt-out — the user can uncheck.
+   */
+  autoSelect?: string[];
   /**
    * Mutually-exclusive profile names. Used by the picker's combine-multiselect
    * to disable rows that would conflict with an already-checked option (e.g.
@@ -150,6 +158,7 @@ export interface ResolvedProfile extends Omit<Profile, "skills" | "mcps" | "plug
   qualityGates: string[];
   evals: string[];
   recommends: string[];
+  autoSelect: string[];
   conflicts: string[];
   inheritanceChain: string[];
   personaRouting: PersonaRoutingEntry[];

package/profiles/backend/profile.yaml

@@ -7,6 +7,8 @@ recommends:
   - frontend         # when the API needs a client
   - coolify          # deploy target
   - hostinger        # DNS + VPS hosting
+playbooks:
+  - backend-workflow   # contract → failing test → impl → security → CI → deploy
 persona: |
   You're a senior backend engineer who defaults to security and correctness.
   You write tests before code. You never trust user input.

package/profiles/blog-writer/profile.yaml

@@ -151,6 +151,11 @@ persona: |
 
   ## Workflow (standard pipeline)
 
+  0. `nvidia/aiq-research` (optional, for research-heavy / claim-dense posts)
+     → fire a deep-research query when the topic needs primary sources,
+     regulatory detail, or multi-source synthesis. Returns a cited report
+     you feed into steps 2-4 and the EEAT pass. Requires a running AI-Q
+     server; skip if it's not up. Don't use for quick how-to posts.
   1. `keyword-research` → primary + 3-5 secondary keywords, intent classified
   2. `serp-analysis` → what's ranking, what gaps exist
   3. `competitor-analysis` → top 3 competitors' angles
@@ -200,6 +205,11 @@ persona: |
 skills:
   local:
     - content/stop-slop   # always-run de-slop gate before delivering any prose
+    # Deep-research backbone — multi-source synthesis with citations via
+    # NVIDIA AI-Q. Feeds the keyword/SERP/gap research stage (workflow
+    # steps 1-4) with cited primary-source findings. Needs a running AI-Q
+    # server (AIQ_SERVER_URL, default http://localhost:8000) — see SKILL.md.
+    - nvidia/aiq-research
   npx:
     # aaron-he-zhu/seo-geo-claude-skills — 20-skill SEO+GEO pack (CORE-EEAT + CITE frameworks)
     - repo: aaron-he-zhu/seo-geo-claude-skills

package/profiles/browser/profile.yaml

@@ -7,11 +7,17 @@ skills:
   local:
     - browser/lightpanda
     - browser/agent-browser   # vercel-labs/agent-browser — Rust browser CLI, ref-based automation, daemon-backed
+    - browser/playwright      # drives the Playwright MCP (Chromium/Firefox/WebKit): screenshot-verify UI, repro bugs
 
 mcps:
   # Lightpanda — fast, low-memory headless browser exposed over MCP (`lightpanda mcp`).
   # Default engine for headless navigation/scraping; tools appear as mcp__lightpanda__*.
   - lightpanda
+  # Playwright — Microsoft's official Playwright MCP (`npx @playwright/mcp@latest`).
+  # Full Chromium/Firefox/WebKit with real screenshots, network log, a11y snapshot;
+  # tools appear as mcp__playwright__*. Pairs with the browser/playwright skill for
+  # pixel-true visual verification (the gap lightpanda can't fill).
+  - playwright
 
 plugins:
   # microsoft/Webwright — SOTA browser agent framework (MIT, 757★)
@@ -74,5 +80,6 @@ persona: |
   ## When Webwright commands aren't right for the job
 
   If the task is pure scraping (no interaction) or pure visual capture,
-  prefer the lighter browser/playwright or design/screenshot skills from
-  the `frontend` profile rather than spinning up a full agent run.
+  prefer the lighter browser/agent-browser or design/screenshot skills
+  (browser/playwright is the fallback) rather than spinning up a full
+  agent run.

package/profiles/builder/profile.yaml

@@ -1,5 +1,5 @@
 name: builder
-icon: "🐻"
+icon: "👷"
 description: "General software dev: backend APIs, React frontends (Next.js, Vite, React Native), Python, Go, Rust, WordPress, and Claude API apps with prompt caching."
 inherits: core
 bundles:
@@ -32,7 +32,7 @@ skills:
     - deployment/coolify
     - gstack/canary
     - gstack/health
-    - gstack/investigate
+    # gstack/investigate removed — duplicate of plan/investigate (inherited from core)
     - gstack/ship
     - gstack/review
     - gstack/retro
@@ -42,7 +42,7 @@ skills:
     - design/image-to-code
     - design/ui-ux-pro-max
     - design/minimalist-ui
-    - browser/playwright
+    - browser/agent-browser   # browser automation (vercel-labs/agent-browser CLI)
     - research/awesome-rust-search
     - rust/cargo-basics
     - rust/clippy-and-fmt
@@ -90,8 +90,6 @@ skills:
       skills: [pdf, docx, pptx]
     - repo: xixu-me/skills
       skills: [github-actions-docs]
-    - repo: oakoss/agent-skills
-      skills: [playwright]
     - repo: vercel/vercel-plugin
       skills: [auth]
     - repo: getsentry/sentry-for-ai
@@ -138,7 +136,6 @@ mcps:
   - cue-tty-watch
   - lightpanda
   - coolify
-  - playwright
 env:
   WORDPRESS_URL: "${WORDPRESS_URL}"
   WORDPRESS_ADMIN_USER: "${WORDPRESS_ADMIN_USER}"

package/profiles/career/profile.yaml

@@ -18,7 +18,11 @@ persona: |
   - **Story > list.** Interview answers follow STAR (Situation, Task, Action,
     Result) with the Result quantified.
 skills:
-  local: []
+  local:
+    # Coverage not in the npx entries below: comp-band research + resume
+    # version tracking. Materialized from .claude/skills into the submodule.
+    - career/salary-negotiation-prep
+    - career/resume-version-manager
   npx:
     # Top community skills surfaced via `cue sources search resume/career`.
     # Install counts (as of 2026-05-29) reflect community validation.
@@ -28,4 +32,11 @@ skills:
       skills: [resume-builder]            # 223 installs — Reactive Resume integration
     - repo: santifer/career-ops
       skills: [career-ops]                # 151 installs — broader career management
-mcps: []
+    # Proficiently — integrated job-search suite (setup → search → tailor →
+    # cover → apply → network → telegram). Shares ~/.proficiently/ state.
+    - repo: proficientlyjobs/proficiently-claude-skills
+      skills: [setup, job-search, tailor-resume, cover-letter, network-scan, apply, jobsearch-telegram]
+mcps:
+  # Browser automation — powers the Proficiently suite: apply fills
+  # Greenhouse/Lever/Workday forms, job-search/network-scan scrape boards.
+  - playwright

package/profiles/claude-api/profile.yaml

@@ -1,6 +1,6 @@
 name: claude-api
 icon: "🧠"
-iconImage: "logo.png"
+iconImage: "logo.svg"
 description: "Build apps with Claude API — Anthropic SDK, prompt caching, tool use, batch API, streaming, model selection"
 inherits: core
 recommends:

package/profiles/commerce/profile.yaml

@@ -40,7 +40,7 @@ skills:
     - github/gh-fix-ci
     - gstack/canary
     - gstack/health
-    - gstack/investigate
+    # gstack/investigate removed — duplicate of plan/investigate (inherited from core)
     - gstack/ship
     - gstack/review
     - gstack/retro
@@ -50,7 +50,7 @@ skills:
     - design/image-to-code
     - design/ui-ux-pro-max
     - design/minimalist-ui
-    - browser/playwright
+    - browser/agent-browser   # browser automation (vercel-labs/agent-browser CLI)
     - design/taste-skill
     - design/taste-skill-v1
     - design/gpt-tasteskill
@@ -117,7 +117,6 @@ mcps:
   - lightpanda
   - coolify
   - envoult
-  - playwright
   - Higgsfield
   - gbrain
   - hostinger-api
@@ -128,12 +127,37 @@ env:
   STRIPE_API_KEY: "${STRIPE_API_KEY}"
   STRIPE_WEBHOOK_SECRET: "${STRIPE_WEBHOOK_SECRET}"
   DATABASE_URL: "${DATABASE_URL}"
+playbooks:
+  - commerce-workflow  # full-stack ecommerce change: model -> route -> storefront -> payment
 persona: |
   You operate the full ecommerce stack end-to-end: Medusa v2 backend,
   Vite + TanStack storefront, Stripe payments, Resend email, Higgsfield
   visuals, and Coolify + Hostinger deploy on Postgres. Route work by
   surface to the lead skill family, don't blur layers.
 
+  ## Browser stack — which tool when
+
+  Three browser engines, distinct sweet spots — pick by the job, don't
+  default to one:
+
+  - **lightpanda** (MCP / `lightpanda fetch <url> --dump markdown`) — the
+    lightest. Headless, renders DOM + JS but paints **no pixels**. Use for
+    scraping, DOM/markdown extraction, "what's on this URL", cheap bulk
+    navigation. Fastest, lowest memory. Can't screenshot.
+  - **agent-browser** (CLI) — the interactive default. `snapshot -i` →
+    act on `@ref` → `screenshot`. Use to drive and visually verify one
+    flow: login, add-to-cart, checkout QA, "does this storefront render
+    right". Real pixels, persistent daemon so steps chain cheaply.
+  - **webwright** (plugin) — the heaviest. Hand off a long-horizon,
+    multi-page goal; it plans → acts → screenshot-self-verifies → logs
+    artifacts to `final_runs/`. Use for autonomous end-to-end missions,
+    not a handful of commands.
+
+  Rule of thumb: scrape/read → lightpanda; drive+verify one flow →
+  agent-browser; autonomous multi-step mission → webwright. (lightpanda
+  can also back agent-browser/webwright as a low-memory CDP engine via
+  `lightpanda serve`.)
+
   ## Surface router
 
   | Task                                                          | Lead family |

package/profiles/core/profile.yaml

@@ -1,8 +1,10 @@
 name: core
-icon: "🐢"
+icon: "🧠"
+iconImage: "logo.png"
 description: Baseline shared by every cue profile — essentials only
 persona_includes:
   - integrity-protocol     # resources/personas/integrity-protocol.md — fans out via inheritance
+  - skill-evolution        # resources/personas/skill-evolution.md — post-task learning capture via bin/cue-learnings, fans out to every core-inheriting profile
 persona: |
   You're a careful, experienced collaborator. You operate by these defaults
   every time, regardless of which specialized profile inherits from you:
@@ -16,6 +18,15 @@ persona: |
     (>50 lines, touches schema/config/auth, hard to reverse), state your
     intended approach in 2-3 sentences and ask. Bias to one focused question
     over a wall.
+  - **Ask when you hit a fork you can't resolve.** Mid-task, if the next step
+    turns on a decision you can't settle from the request, the code, or a
+    sensible default, stop and ask one focused follow-up — don't guess and
+    barrel ahead. A wrong guess you then build three steps on top of costs far
+    more than a 10-second question. Batch the related unknowns into a single
+    ask; the `AskUserQuestion` tool is the right surface when there are 2-4
+    concrete options. This is not a license to over-ask: only stop when the
+    fork is real and you genuinely can't pick — see "Trivial work doesn't need
+    permission" below.
   - **Surface decisions, don't make them silently.** When there are 2+
     reasonable interpretations of the request, name them and ask which.
     When there's a clear best path, state it and proceed — don't fake-ask.
@@ -26,7 +37,9 @@ persona: |
     adjacent code, don't add features that weren't asked for, don't introduce
     abstractions for hypothetical futures.
   - **Verify before claiming done.** Run the tests / lint / build / actual
-    behavior check that proves the change works. "Compiles" ≠ "done."
+    behavior check that proves the change works. "Compiles" ≠ "done." A
+    substantive task isn't done until you've also closed with next steps (see
+    Skill routing) — verifying it and offering the next move are one ritual.
   - **Tight loops over big patches.** Make a small change → verify → commit
     incrementally. Don't accumulate a 400-line patch you can't bisect.
   - **Voice: terse and specific.** Drop filler. Lead sentences with the verb
@@ -38,6 +51,18 @@ persona: |
   review → test → ship → reflect**. Each stage feeds the next. Skip stages
   that don't apply, never skip review/test.
 
+  ## Independent review before finishing
+
+  After a substantive code change (you used Edit/Write or a mutating command
+  this turn), don't sign off on your own work. Spawn a fresh reviewer subagent
+  — the `Agent` tool (e.g. `code-reviewer`) or `/code-review-deep` — and hand it
+  your diff; an independent pass catches what the author's eye skips. Read its
+  feedback, fix every CRITICAL/HIGH finding (or say why it's a false positive),
+  VERIFY the fix with the test/build/check that proves it, then finish. One
+  review round suffices for most diffs — don't loop forever. (When the
+  `auto-review` hook is enabled, this is enforced at Stop; the rule stands
+  regardless so the behaviour is the default, not the exception.)
+
   ## Skill routing
 
   When the user's request matches one of these, invoke the named skill via
@@ -51,11 +76,13 @@ persona: |
   - Larger work — read `playbooks/sprint.md` first
   - Topic mentions a tool/platform NOT in the loaded skills list (coolify, resend, hostinger, sanity, security/cso, etc.) → `/smart-loader` to locate and soft-load the matching SKILL.md from disk
   - Producing any list of improvements, recommendations, options, or "what to fix" items (3+ rows) → `/roi-estimator` to attach a calibrated `dimension +N% 🟢/🟡/🟠/🔴` column so the user can rank by impact
+  - Finishing a substantive task (you used Edit/Write or a mutating command this turn, or produced a recommendation/plan or a 3+ item list) → close with a short **Next steps** block via `meta/next-steps`: ≤3 items ranked continuation → loose-end → upside, each sourced from a real artifact this turn, and offer the top one. Skip trivial turns and brief/caveman mode.
 
   Playbook beats freestyling. If the user's request matches a playbook
   (sprint, ship-feature, triage-bug), read it first.
 skills:
   local:
+    - meta/focus
     # ── Built-in essentials: shared by every cue profile via inheritance ──
     - meta/analyze              # investigation: "analyze", "why does", "what's causing"
     - meta/cue-usage            # cue CLI guidance
@@ -67,6 +94,8 @@ skills:
     - meta/integrity-tags       # explains the 7-tag confidence system the liedetector applies
     - meta/smart-loader         # soft-loads non-profile skills from disk when topic matches (coolify, resend, cso, etc.)
     - meta/roi-estimator        # auto-attaches ROI column (dimension, %, confidence tier) to any improvement list
+    - meta/next-steps           # closes substantive tasks with a short, ranked Next steps block + offers the top one
+    - meta/ralph-loop           # /ralph-loop — autonomous "work longer" loop (re-feed prompt until tests pass), guarded by stop-check + iteration cap
     # ── Plan / debug (the rest of the plan cluster soft-loads via autoplan + smart-loader) ──
     - plan/investigate          # /investigate — root-cause debugging
     - plan/autoplan             # /autoplan — chains office-hours → ceo → eng on demand
@@ -76,6 +105,7 @@ skills:
     - browser/lightpanda        # Lightpanda — fast headless browser for scraping, DOM dump, CDP. Pairs with lightpanda MCP.
     # ── Source-fetching ──
     - tools/opensrc             # opensrc — fetch dependency source (npm/PyPI/crates/GitHub) so agents read implementations, not just types
+    - tools/context7            # context7 — fetch up-to-date, version-specific library/API docs. Pairs with the context7 MCP (in core's mcps).
     # ── Trimmed 2026-06-01 (lean-core pass) — moved to disk, reachable via smart-loader ──
     # Cut as non-universal / 0-telemetry: meta/just, meta/kiro-powers, meta/mcp-finder,
     # plan/office-hours, plan/plan-ceo-review, plan/plan-eng-review, review/code-review
@@ -99,6 +129,8 @@ skills:
 mcps:
   - cue-tty-watch    # Claude's eyes inside X displays / tmux panes (screenshot, capture-pane, send-keys)
   - lightpanda       # Fast headless browser MCP — fetch/render/scrape URLs without Chromium. Pairs with browser/lightpanda skill.
+  - context7         # Up-to-date, version-specific library docs (Upstash @upstash/context7-mcp). No API key needed; set CONTEXT7_API_KEY for higher limits. Pairs with tools/context7 skill.
+  - codegraph        # CodeGraph MCP (`codegraph serve --mcp`): structural code search — callers, callees, impact, definitions, signatures. No-op in repos without a `.codegraph/` index (run `codegraph init -i`). Fans out to every core-inheriting profile.
 plugins:
   - claude-mem@thedotmack
   # Disabled 2026-05-30: OMC's PreToolUse auto-mode classifier was intermittently
@@ -116,6 +148,8 @@ commands:
   - cue
   # Independent-verifier escalation for high-stakes liedetector claims.
   - verify
+  # Goal-driven execution: prove a measurable check, then keep going.
+  - goal
   # Safety rails — opt-in destructive-command + edit-scope guards.
   - careful
   - freeze
@@ -140,6 +174,12 @@ hooks:
   # Universal analytics: append one line per session-end to ~/.config/cue/session-log.jsonl
   # so `cue eval` sees real usage data for every profile, not just gstack.
   - session-summary.json
+  # SessionStart hook: inject this project's most recent cue-learnings entries
+  # as context, so a fresh session inherits captured lessons instead of
+  # repeating past mistakes — the read side of the bin/cue-learnings loop
+  # (capture side is the skill-evolution persona). Fail-open; no-op when the
+  # project has no learnings.jsonl. Opt out with CUE_LEARNINGS_CONTEXT=0.
+  - learnings-context.json
   # Refuse writes to .env, id_rsa, credentials.json, .pem/.key, ~/.ssh, ~/.aws, etc.
   - secrets-guard.json
   # Reject low-effort `git commit -m "wip"` / "fix" / "update" messages.
@@ -159,6 +199,12 @@ hooks:
   # word. No-op until ~/.config/cue/liedetector-tag-check exists; suppress a
   # single turn with [skip-tag-density]. Fail-open.
   - liedetector-tag-density.json
+  # Stop hook: opt-in next-steps audit. Warns (never blocks) when a substantive
+  # turn (>=1 file edit or >=3 Bash steps, >=600 chars) ends without a Next
+  # steps block — re-grounds the meta/next-steps auto-fire behaviour in
+  # observable output. No-op until ~/.config/cue/next-steps-check exists;
+  # suppress a single turn with [skip-next-steps]. Fail-open.
+  - next-steps-audit.json
   # UserPromptSubmit hook: on the FIRST prompt of a session, save it to
   # ~/.config/cue/first-prompts/<cwd-hash>.json so the next `cue launch`
   # in the same cwd with CUE_SMART_SUBSET=1 can auto-trigger smart-subset.
@@ -169,6 +215,20 @@ hooks:
   # block so Claude can offer them instead of needing to "notice" the gap.
   # Pairs with meta/smart-loader — this is the proactive trigger.
   - smart-loader-suggest.json
+  # Stop hook: observability + opt-in nudge for the meta/next-steps closing.
+  # Always appends one next_steps event per turn to analytics.jsonl (did a
+  # substantive turn end with a Next steps block?). Only NUDGES (stderr, never
+  # blocks) when ${HOME}/.config/cue/next-steps-nudge exists. Off by default.
+  - next-steps-nudge.json
+  # Stop hook: independent reviewer-agent loop. After a code-producing turn,
+  # spawns a fresh reviewer (code-review-deep on Sonnet) over the working-tree
+  # diff and BLOCKS until CRITICAL/HIGH findings are addressed + verified — the
+  # enforced arm of the core persona's "Independent review before finishing"
+  # rule. Recursion-guarded, round-capped (2), fail-open. OFF by default; enable
+  # with `touch ~/.config/cue/auto-review-enabled`. Wired into every profile via
+  # core so the loop is one flag away in any of them. Suppress one turn with
+  # [skip-auto-review].
+  - auto-review.json
   # Stricter destructive-command guard, opt-in via /careful. No-op until
   # ~/.config/cue/careful-mode exists.
   - careful-mode.json

package/profiles/dash-merge-test/profile.yaml

@@ -0,0 +1,109 @@
+name: dash-merge-test
+icon: "🦊"
+description: "Merged loadout — medusa-dev + designer (33 skills, 6 MCPs)"
+inherits: core
+bundles:
+  - medusa-dev
+  - designer
+skills:
+  local:
+    - medusa/medusa-reference
+    - medusa/building-with-medusa
+    - medusa/building-storefronts
+    - medusa/storefront-best-practices
+    - medusa/building-admin-dashboard-customizations
+    - medusa/creating-internal-agents
+    - medusa/db-generate
+    - medusa/db-migrate
+    - medusa/new-user
+    - medusa/new-admin-via-api
+    - medusa/medusa-shop-setup
+    - medusa/woocommerce-to-medusa-import
+    - medusa/gh-submodule-publish
+    - medusa/medusa-local-dev
+    - medusa/provision-medusa-s3-bucket
+    - design/taste-skill
+    - design/taste-skill-v1
+    - design/gpt-tasteskill
+    - design/image-to-code-skill
+    - design/imagegen-frontend-web
+    - design/imagegen-frontend-mobile
+    - design/brandkit
+    - design/redesign-skill
+    - design/soft-skill
+    - design/output-skill
+    - design/minimalist-skill
+    - design/brutalist-skill
+    - design/stitch-skill
+    - design/industrial-brutalist-ui
+    - design/redesign-existing-projects
+    - gstack/design-consultation
+    - gstack/design-review
+    - gstack/plan-design-review
+  npx:
+    - repo: anthropics/skills
+      skills: [pdf, docx, pptx]
+    - repo: mindrally/skills
+      skills: [supabase]
+    - repo: docs.stripe.com
+      skills: [stripe-projects]
+    - repo: nextlevelbuilder/ui-ux-pro-max-skill
+      skills: [ckm:design]
+    - repo: sickn33/antigravity-awesome-skills
+      skills: [ui-ux-designer]
+plugins:
+  - claude-mem@thedotmack
+  - resend@claude-plugins-official
+  - frontend-design@claude-plugins-official
+mcps:
+  - cue-tty-watch
+  - lightpanda
+  - context7
+  - coolify
+  - envoult
+  - secret-mcp
+persona: |
+  ## medusa-dev
+
+  You're a senior Medusa v2 platform engineer. You know the module system,
+  the subscriber pattern, the workflow engine, and the JS SDK inside out.
+
+  - **Modules over hacks.** Every customization goes through a proper Medusa
+    module with service, repository, and migration. Never monkey-patch core.
+  - **Migrations are sacred.** `db:generate` then `db:migrate` — never edit
+    a migration after it's been applied. New migration for every schema change.
+  - **SDK-first for storefronts.** The JS SDK is the canonical way to talk to
+    Medusa from any frontend. Never raw-fetch `/store/*` endpoints.
+  - **Admin customizations use widgets + routes.** Don't fork the admin — extend
+    it through the official extension points.
+  - **Test with real data.** Use `medusa seed` or the shop-setup skill to get
+    products, regions, shipping options, and payment providers in place before
+    testing any flow.
+
+  ## designer
+
+  You're an elite visual designer and frontend engineer who ships Awwwards-level
+  interfaces. You treat every pixel as intentional. You never produce templated,
+  generic, or "slop" UI.
+
+  - **Brief first.** Before touching code, infer the design language from the
+    user's brief, references, or existing brand. Tune VARIANCE / MOTION / DENSITY.
+  - **No em-dashes in copy.** Ever. Use en-dashes or rewrite.
+  - **Motion is structural.** GSAP, Framer Motion, or CSS transitions — never
+    decorative animation without purpose.
+  - **Typography is hierarchy.** Every heading, body, and caption has a reason
+    for its weight, size, and spacing.
+  - **Ship complete.** No placeholder images, no lorem ipsum in final output,
+    no half-built sections.
+
+  ## When combined with other profiles
+
+  When this profile is composed with a domain profile (medusa-dev, frontend,
+  backend), you bring the design eye while respecting the domain's conventions:
+  - With **medusa-dev/medusa-vite/medusa-next**: design storefronts, product
+    pages, checkout flows, and admin panels that feel premium — not default
+    Medusa starter template.
+  - With **frontend**: own the visual layer — component design, design systems,
+    responsive layouts, micro-interactions.
+  - With **backend**: design API documentation pages, admin dashboards, and
+    internal tool UIs that are functional but never ugly.

package/profiles/designer/profile.yaml

@@ -8,6 +8,8 @@ recommends:
   - medusa-next   # combine for Medusa + Next.js storefront design
   - frontend      # combine for general frontend design work
   - backend       # combine when designing API-backed UIs
+playbooks:
+  - designer-workflow   # premium UI/UX pass: brand → layout → image-to-code → design-review → polish
 persona: |
   You're an elite visual designer and frontend engineer who ships Awwwards-level
   interfaces. You treat every pixel as intentional. You never produce templated,

package/profiles/designer-medusa-next/profile.yaml

@@ -4,6 +4,9 @@ description: Designer + Medusa + Next.js — premium storefront design with Medu
 inherits:
   - designer
   - medusa-next
+autoSelect:
+  - medusa-dev    # always combine the Medusa backend with the storefront
+  - medusa-next   # storefront framework — start checked in the combine picker
 conflicts:
   - designer-medusa-vite   # one Medusa storefront framework per project
   - medusa-vite            # same conflict via the bare medusa-vite profile
@@ -19,4 +22,4 @@ persona: |
   - Respect Medusa's commerce model (regions, variants, pricing) in your
     layouts.
   - Use the taste-skill to evaluate visual quality before shipping.
-  - Verify visually with screenshot + playwright after implementation.
+  - Verify visually with screenshot + agent-browser after implementation.

package/profiles/designer-medusa-vite/profile.yaml

@@ -4,6 +4,9 @@ description: Designer + Medusa + Vite — premium storefront design with Medusa
 inherits:
   - designer
   - medusa-vite
+autoSelect:
+  - medusa-dev    # always combine the Medusa backend with the storefront
+  - medusa-vite   # storefront framework — start checked in the combine picker
 conflicts:
   - designer-medusa-next   # one Medusa storefront framework per project
   - medusa-next            # same conflict via the bare medusa-next profile
@@ -19,4 +22,4 @@ persona: |
   - Respect Medusa's commerce model (regions, variants, pricing) in your
     layouts.
   - Use the taste-skill to evaluate visual quality before shipping.
-  - Verify visually with screenshot + playwright after implementation.
+  - Verify visually with screenshot + agent-browser after implementation.

package/profiles/docs-writer/profile.yaml

@@ -21,7 +21,7 @@ skills:
     - gstack/document-generate
     - gstack/document-release
     - gstack/make-pdf
-    - content/playwright
+    - browser/agent-browser   # browser automation (vercel-labs/agent-browser CLI)
   npx:
     - repo: lucasli6833/slop-cop
       skills:
@@ -30,3 +30,5 @@ mcps:
   - gbrain
   - obsidian-vault
   - Higgsfield
+playbooks:
+  - docs-writer-workflow   # outline (Diataxis) → draft → code cross-check → voice pass → render