bmad-plus 0.21.0 → 0.23.0

package/readme-international/README.fr.md

@@ -4,9 +4,9 @@
   <a href="../README.md">English</a> | 🌐 <b>Français</b> | <a href="README.es.md">Español</a> | <a href="README.de.md">Deutsch</a>
 </div>
 
-[![Version](https://img.shields.io/badge/version-0.21.0-blue)](https://www.npmjs.com/package/bmad-plus)
+[![Version](https://img.shields.io/badge/version-0.23.0-blue)](https://www.npmjs.com/package/bmad-plus)
 
-**Version 0.21.0** · Node.js `>=20.0.0` · MIT
+**Version 0.23.0** · Node.js `>=20.0.0` · MIT
 
 Des workflows de développement IA propres à chaque projet, avec des rôles clairs, un contexte partagé, des mises à jour sûres et des adaptateurs pour vos outils de code.
 
@@ -39,7 +39,7 @@ Lancez l’installateur. Sélectionnez les adaptateurs de vos outils IA et les p
 **Dans le terminal du projet :**
 
 ```sh
-npx bmad-plus@0.21.0 install
+npx bmad-plus@0.23.0 install
 ```
 
 L’installateur crée les instructions des agents, le socle partagé et les adaptateurs sélectionnés. Certains packs facultatifs nécessitent d’autres environnements ou des accès API.
@@ -102,6 +102,7 @@ Relisez les changements et les résultats des tests. L’assistant suit les auto
 | Dev Studio | Des rôles et workflows spécialisés en produit, design, ingénierie et documentation. |
 | Backup | Des instructions et utilitaires pour les sauvegardes horodatées, la restauration et la rotation. |
 | Animated | Un workflow dédié à la création de sites dont le défilement s’appuie sur la vidéo. |
+| Design | Atelier, le moteur de design : construire, auditer et refondre des interfaces selon des règles vérifiables, interview d’abord. |
 
 ## Exemples
 
@@ -175,18 +176,27 @@ Lis AGENTS.md et la mémoire de projet disponible. Résume le dernier état vér
 
 Résultat attendu : une reprise courte et fondée sur des éléments vérifiés. Les notes apportent du contexte, à confronter au travail actuel.
 
-## Nouveautés de la 0.21.0
+### Concevoir un écran avec Atelier
 
-Des revues qui s’arrêtent honnêtement, reprennent et parlent à la CI
+À utiliser avec le pack Design facultatif installé. Atelier pose d’abord ses questions, puis suit des règles vérifiables.
 
-Une revue enregistre désormais pourquoi elle s’est arrêtée et ce qu’elle a consommé, écrit un contrôle lisible par une étape de CI, et reprend sur le périmètre qu’elle a scellé. Le récupérateur SEO ne se connecte qu’aux adresses qu’il a validées, et le serveur MCP passe au SDK officiel sous Python 3.12. Rien ne change dans l’installation ou la mise à jour.
+```text
+Atelier, fais la lecture design de la liste des factures de mon application, construis l’écran et termine par le verdict de pré-vol.
+```
+
+Résultat attendu : la lecture design enregistrée, l’écran avec tous ses états couverts, et le verdict de pré-vol cité en dernière ligne.
+
+## Nouveautés de la 0.23.0
+
+Demander d’abord, puis lancer l’essaim
+
+Un pack déclare désormais les questions auxquelles il lui faut une réponse avant d’agir, et Nexus peut lancer un essaim de sous-agents, un panel simulé ou un jury, dans un budget que vous fixez. Atelier, le moteur de design, devient le dixième pack. Installation et mise à jour ne changent pas.
 
-- coverage.json peut enregistrer le déroulé d’une revue : pourquoi elle s’est arrêtée (terminée, budget, limite de temps, trois échecs de suite, interrompue), les passes faites et l’usage rapporté par l’hôte. Une revue arrêtée avant la fin est toujours incomplète, et une unité est abandonnée après trois échecs consécutifs.
-- bmad-plus review gate --emit-check suivi d’un nom de fichier écrit le verdict sous forme de contrôle pour la CI : disposition, raisons, remarques ouvertes par gravité avec leurs lignes, et un bloc au format des check runs GitHub. Les mêmes preuves donnent toujours le même fichier.
-- bmad-plus review continue reprend une revue interrompue sur le même périmètre scellé et liste ce qui reste dû. Il refuse quand le code, les règles ou un fichier retenu ont changé, et renvoie vers un nouveau périmètre et review compare.
-- Les règles de revue peuvent appartenir à un groupe (code, data, delivery ou le vôtre), pour que des relecteurs parallèles se partagent une unité par famille de règles.
-- Pour les mainteneurs, un banc d’essai note des réponses de revue enregistrées face à des défauts plantés et des leurres, avec une règle de correspondance publiée, et des modèles de CI prêts à copier font passer une pull request par un agent hôte sur un runner auto-hébergé sans rien croire de ce que l’agent peut réécrire.
-- Le récupérateur et le crawler SEO résolvent eux-mêmes chaque hôte, refusent les adresses internes et se connectent exactement à l’adresse validée, redirections comprises, ce qui neutralise le DNS rebinding. Le serveur MCP utilise le SDK officiel sous Python 3.12 avec chromadb et sentence-transformers à jour ; relancez ingest.py une fois après la mise à jour.
+- bmad-plus intake pose les questions d’un pack par petits lots, marque les indispensables, propose des valeurs par défaut, note ce que vous ignorez avec le risque associé, et rédige le brief qui devient le contrat de tout ce que le pack fait ensuite. Rien n’y est inventé.
+- bmad-plus swarm tire des unités simulées à partir d’un profil sous une graine, écrit un brief par sous-agent, valide chaque fichier de réponse, ne compte que les réponses valides, et rappelle que des acheteurs simulés ne sont pas des clients. Le bloc swarm de _bmad/config.yaml fixe le nombre de sous-agents lancés avant que l’assistant ne vous demande ; la génération de médias payante demande toujours.
+- Le registre enregistre désormais quelles compétences un agent pilote, de quel paquet npm viennent les compétences d’un pack, quelles étapes de pipeline et quelles portes un pack déclare et quel questionnaire il joue. Les instructions générées portent tout cela, rien n’est tapé deux fois.
+- Pack Design : Atelier, construit à partir du paquet design-skill, vous interroge avec une lecture design, puis construit, audite ou refond une interface selon des règles vérifiables. L’installateur copie ses trois compétences et leurs modules de référence, et doctor les vérifie.
+- Une porte de déclencheurs refuse les règles d’activation automatique qui nomment un agent ou un pipeline inconnu du registre, et l’installateur n’écrit rien quand les fichiers d’un pack ne peuvent pas tous être listés d’abord.
 
 ## Historique des versions
 
@@ -194,9 +204,9 @@ Ces dates identifient les notes relues du CHANGELOG, pas les dates de publicatio
 
 | Version | Date des notes | Résumé relu |
 | --- | --- | --- |
+| 0.23.0 | 2026-10-17 | Un pack déclare désormais les questions auxquelles il lui faut une réponse avant d’agir, et Nexus peut lancer un essaim de sous-agents, un panel simulé ou un jury, dans un budget que vous fixez. Atelier, le moteur de design, devient le dixième pack. Installation et mise à jour ne changent pas. |
+| 0.22.0 | 2026-09-29 | BMAD+ déclare désormais chaque envoi de données et chaque processus qu’il lance, et sa CI refuse tout ce qui n’est pas déclaré. La page de recette et le rapport SEO ne chargent plus rien depuis d’autres sites, les appels OSINT passent uniquement en HTTPS, et Shield gagne un dossier de sécurité fondé sur des preuves exécutées et un registre des traitements IA. Rien ne change dans l’installation ou la mise à jour. |
 | 0.21.0 | 2026-09-28 | Une revue enregistre désormais pourquoi elle s’est arrêtée et ce qu’elle a consommé, écrit un contrôle lisible par une étape de CI, et reprend sur le périmètre qu’elle a scellé. Le récupérateur SEO ne se connecte qu’aux adresses qu’il a validées, et le serveur MCP passe au SDK officiel sous Python 3.12. Rien ne change dans l’installation ou la mise à jour. |
-| 0.20.0 | 2026-09-28 | bmad-plus review donne désormais au relecteur les règles qui s’appliquent aux fichiers modifiés, fixe la profondeur d’une revue avant qu’elle commence, compare une revue avec une précédente sans déclarer corrigée une remarque non vérifiée, et masque les identifiants dans ce qu’il écrit. Rien ne change dans l’installation ou la mise à jour. |
-| 0.19.0 | 2026-09-27 | bmad-plus review scelle ce qu’une revue doit couvrir, retrouve chaque remarque par le code qu’elle cite, et déduit le verdict de la couverture plutôt que du nombre de remarques. Le tri des revues garde désormais une remarque tant qu’un motif cité ne la réfute pas. Rien ne change dans l’installation ou la mise à jour. |
 
 [Historique des versions et guide de mise à jour](https://bmad-plus.rochetta.fr/fr/docs/#changelog) · [Toutes les versions publiées sur npm](https://www.npmjs.com/package/bmad-plus?activeTab=versions)
 

package/src/bmad-plus/agents/agent-orchestrator/SKILL.md

@@ -73,6 +73,8 @@ authorization and completed work when changing depth.
 
 Read `execution_mode` and `checkpoints` from `_bmad/config.yaml` before starting the pipeline. The checkpoint keys are `discovery`, `architecture`, `story`, and `delivery`.
 
+**Pack pipelines.** When the request or the context names a pack that declares `pipelines` in the registry (the spine's facts section counts them on the pack's line and lists each pipeline's steps on a `- Pipelines (<pack>)` line; read the steps there), run that pipeline instead of the generic phases for that pack: its steps in order, each step's `workflow` or `agent` from the pack, each step's `gate` applied with the checkpoint behavior below. When the pack also declares an `intake`, run it first with the `bmad-plus-intake` skill and do not start the first step while a must-have is neither answered nor marked unknown with its risk; the rendered brief is the input of every step.
+
 **Checkpoint behavior** (configurable):
 - `require_approval` (🔴 STOP): Check whether the existing user authorization covers the concrete next action. If approval is still required, present the reviewable checkpoint and wait; do not ask again for an action already authorized. If an external channel is configured and sending is authorized, also notify through it.
 - `notify_only` (🟡 INFO): Report progress in the session and continue. If an external channel is configured and sending is authorized, also notify through it.
@@ -101,7 +103,9 @@ Focuses on: detecting parallelizable tasks, launching agents in parallel, confli
 - **Reallocate**: Agent finishes early → assign next available task
 - **Escalate**: A task exhausts its attempt limit or cannot be reconciled → report the condition, current evidence and preserved changes. Continue independent authorized work when possible. An external notification still needs configuration and authorization.
 
-> 💡 **Auto-activates** when: multiple independent stories are queued, or when `{parallel_execution}` is enabled and batch work is detected.
+**Swarms.** A pack that ships a swarm definition (a panel of simulated units, a jury of lenses, a fanout of disjoint scopes) is run through `bmad-plus swarm` and the `bmad-plus-swarm` skill, one wave of session sub-agents at a time, each reading its brief as its whole instruction; a fanout can instead be exported as a Nexus plan for durable host attempts. Stay within `swarm.budget` from `_bmad/config.yaml`; beyond it, ask the user before `init --approved`. Paid media generation is never covered by that budget.
+
+> 💡 **Auto-activates** when: multiple independent stories are queued, when `{parallel_execution}` is enabled and batch work is detected, or when a pack's swarm is declared or a panel is requested.
 
 When auto-activating a role, **announce it**: "💡 I'm switching to [Role] mode — [reason]. Say 'skip' to stay in current mode."
 

package/src/bmad-plus/data/role-triggers.yaml

@@ -211,6 +211,7 @@ orchestrator:
       contexts:
         - "new_project_idea"
         - "execution_mode_autopilot"
+        - "pack_pipeline_declared"
 
     parallel-supervisor:
       patterns:
@@ -223,6 +224,8 @@ orchestrator:
         - "independent_stories_detected"
         - "multi_task_queue"
         - "parallel_execution_enabled"
+        - "swarm_declared"
+        - "panel_requested"
       reasoning:
         - "stories_have_no_shared_dependencies"
         - "workload_can_be_distributed"
@@ -278,3 +281,34 @@ seo:
         - "competitive_gap_analysis"
       reasoning:
         - "findings_need_prioritized_scoring"
+
+design:
+  default_role: design-engine
+  auto_triggers:
+    design-engine:
+      # Level 1 — Pattern matching (keywords in user request). The persona picks its
+      # role (Builder, Reviewer, Redesigner) from the request; the pack has one agent.
+      # The vocabulary is the pack's own: bare "design", "UI" or "interface" belong to
+      # Forge's architect role and Sentinel's UX review, which coexist with this pack.
+      patterns:
+        - "Atelier"
+        - "/design"
+        - "design tokens"
+        - "maquette"
+        - "wireframe"
+        - "landing page"
+        - "make it look good"
+        - "accessibilité"
+        - "accessibility"
+        - "refonte"
+        - "redesign"
+      # Level 2 — Contextual (situations detected during work)
+      contexts:
+        - "frontend_changes"
+        - "new_page"
+        - "new_component"
+        - "design_tokens_missing"
+      # Level 3 — Reasoning chain (discoveries during execution)
+      reasoning:
+        - "ui_looks_generic"
+        - "accessibility_unverified"

package/src/bmad-plus/module-help.csv

@@ -9,3 +9,5 @@ bmad-plus-agent-shadow,"Talk to Shadow — your OSINT Intelligence Analyst. Inve
 bmad-plus-sync,"Synchronize BMAD+ with upstream BMAD-METHOD. Check for updates, analyze changes via AI, and merge compatible updates. Uses VPS MCP Server for git operations."
 bmad-plus-agent-maker,"Talk to Maker — your Agent Creator. Designs, builds, and packages new BMAD+ compatible agents from a natural language description. Outputs ready-to-integrate packages."
 bmad-plus-uat,"Human acceptance recipes (recette): turn a delivery into a step-by-step page a person plays, read their run, classify every failure, and gate the delivery on it."
+bmad-plus-intake,"Pack intake: ask the questions a pack needs before it acts, in the user's language and a few at a time, record each answer, default or unknown, and render the brief the pack works from."
+bmad-plus-swarm,"Run a swarm of sub-agents under a budget: deal simulated buyers or a jury, brief one sub-agent per unit, file and check their answers, tally them with the disclosure first. Use for panels, juries and fanouts."

package/src/bmad-plus/module.yaml

@@ -35,6 +35,10 @@ install_packs:
       label: 🗂️ Universal Backup — Timestamped ZIP backup with smart exclusions.
     - value: animated
       label: 🎬 Animated Website — Luxury scroll-driven website from video.
+    - value: design
+      label: >-
+        🎨 Design — Atelier builds, audits and redesigns interfaces under checkable rules: intake first,
+        semantic tokens, WCAG AA, every state covered, pre-flight before delivery.
     - value: all
       label: Tout installer
     - value: none
@@ -116,8 +120,10 @@ packs:
       - agent-orchestrator
     skills:
       - bmad-plus-autopilot
+      - bmad-plus-intake
       - bmad-plus-parallel
       - bmad-plus-sync
+      - bmad-plus-swarm
       - bmad-plus-uat
     data:
       - role-triggers.yaml
@@ -332,3 +338,18 @@ packs:
     skills: []
     packDir: pack-animated
     packSrcDir: packs
+  design:
+    name: Design
+    icon: 🎨
+    description: >-
+      Atelier builds, audits and redesigns interfaces under checkable rules: intake first, semantic tokens,
+      WCAG AA, every state covered, pre-flight before delivery.
+    required: false
+    agents:
+      - design-engine
+    skills: []
+    cohabitation_warning: >-
+      Overlaps Sentinel's UX-review role and Dev Studio's Rachel (UX Designer); when this pack is installed
+      its rules govern visual design and accessibility, and the others hand those over.
+    packDir: pack-design
+    packSrcDir: packs

package/src/bmad-plus/packs/pack-design/SKILL.md

@@ -0,0 +1,72 @@
+---
+name: design-engine
+description: >
+  BMAD+ design pack. Atelier plans, builds, audits and reworks interfaces by opening the
+  installed design, design-review and redesign skills after the pack's intake. Use when the
+  user says /design or Atelier, asks for a maquette, a wireframe, a landing page, design
+  tokens, an accessibility review, or the redesign of a live site.
+---
+
+# Design pack — dispatcher
+
+> By Laurent Rochetta | BMAD+ design pack
+
+## Where the rules are
+
+This pack owns no design rule. The rules are the three skills of the `design-skill` package
+(Atelier), copied by the installer next to this pack:
+
+| Skill | Installed at | Owns |
+|---|---|---|
+| `design` | `.agents/skills/design/SKILL.md` | what a new screen must satisfy, the reference modules per surface, the delivery checklist |
+| `design-review` | `.agents/skills/design-review/SKILL.md` | the audit protocol and its severity scale |
+| `redesign` | `.agents/skills/redesign/SKILL.md` | how a live product is changed without losing what works |
+
+This pack adds three things only: the **intake** (`intake.yaml`, the questions Atelier needs
+answered before it acts), the **routing** between the three skills, and the **handoffs** to the
+other BMAD+ agents. The persona that drives them is `design-engine.md`, next to this file.
+
+## Missing installation
+
+Before any command, check that `.agents/skills/design/SKILL.md` exists. When it does not, the
+pack was installed without its skills (the `design-skill` dependency did not resolve, or the
+copy was interrupted). Report it, point at `bmad-plus doctor`, which names the missing files,
+and stop. Do not improvise design rules, do not paraphrase them from memory, and do not treat
+this dispatcher as a design manual: without its skills it is a router with nothing to route to.
+
+## Command router
+
+Every command starts by loading `design-engine.md`. The intake runs, or resumes, whenever
+`bmad-plus intake plan` on `_bmad-output/design/intake.json` is not ready, or when
+`intake.yaml` changed since the answers were recorded; `_bmad-output/design/brief.md` is
+rendered again from the answers each time.
+
+| Command | Role | Action |
+|---|---|---|
+| `/design read` | Builder | Run or resume the intake with the `bmad-plus-intake` skill, settle the dials and the stack, render `_bmad-output/design/brief.md`, print the one-line design read (surface · audience · aesthetic family) and the three dials with their origin (set or inferred) |
+| `/design build <surface>` | Builder | Brief ready → open `design`, follow its routing for `<surface>`, build, deliver with the skill's checklist verdict quoted |
+| `/design review <path or url>` | Reviewer | Open `design-review`, audit the given files or page, write `_bmad-output/design/review.md` as the skill prescribes, every finding labelled `tool` or `inspection` |
+| `/design redesign` | Redesigner | Open `redesign` and follow it end to end: audit, decision with its reason, migration plan with the search and accessibility risks it flags, build through `design`, verification against the old version |
+| `/design tokens` | Builder | Open `design` and its token modules; produce the token sheet for the stack the brief records |
+
+`/design` alone, or `/design help`, prints this table and the state of the intake.
+
+## What this pack does not do
+
+- It runs no script and starts no process. Checks that need a browser or a scanner run only
+  through the project's own tooling, on the user's say-so, and the report labels a tool run
+  apart from an inspection.
+- It sends nothing anywhere. BMAD+ copies the `skills/` directory of the `design-skill`
+  package and nothing else; the package also ships a linker CLI (`bin/atelier.mjs`) that
+  BMAD+ never invokes, and nothing in the copied files runs.
+- It does not decide for Sentinel, the SEO pack or Shield. The handoff table in
+  `design-engine.md` says what each receives and when.
+
+## Files of this pack
+
+| File | Role |
+|---|---|
+| `SKILL.md` | this dispatcher |
+| `design-engine.md` | the Atelier persona: identity, the BMAD+ rules, activation, roles, capabilities, handoffs |
+| `intake.yaml` | the design read as an intake (`bmad-plus intake check` validates it) |
+| `bmad-skill-manifest.yaml` | the pack manifest read by the skill catalog |

package/src/bmad-plus/packs/pack-design/bmad-skill-manifest.yaml

@@ -0,0 +1,13 @@
+type: pack
+name: bmad-plus-pack-design
+displayName: Design Pack (Atelier)
+title: Design — Atelier
+icon: "🎨"
+capabilities: "interface planning, screen building, marketing pages, dashboards, record tables, forms, sign-in and first-run flows, mobile views, HTML email, token sheets, colour and type choices, spacing, motion with reduced-motion paths, accessibility audits, state coverage, interface review, rework of live products, component library choice, handoff to QA and SEO"
+role: "Interface agent of BMAD+: plans, builds, audits and reworks screens by opening the installed design, design-review and redesign skills once the pack intake is complete"
+identity: "Works from a brief, not from a prompt. Asks what the page must do and for whom before choosing a layout, opens the installed skills for the rules instead of recalling them, and hands a screen back when the skill's delivery checklist is not met. Keeps the status of every intake answer visible in what it delivers."
+communicationStyle: "Short, concrete, file-first: names the file, the rule and the fix. Orders a report by what blocks release, then what confuses, then what polishes. Says what a tool measured and what was read against the checklist, and keeps the two apart. French with the user, English in code and technical notes."
+principles: "Intake before layout. The brief keeps the status of every answer. The installed skills own the design rules; this pack routes to them and adds nothing of its own. A report never mixes a tool result with an inspection. Scope is the request, not the whole product. Other BMAD+ agents receive a named handoff, not a hint."
+module: bmad-plus
+canonicalId: bmad-plus-pack-design
+pack: design

package/src/bmad-plus/packs/pack-design/design-engine.md

@@ -0,0 +1,126 @@
+# Atelier — Design Engine
+
+## Identity
+
+You are **Atelier**, the BMAD+ agent for interfaces. You plan, build, audit and rework the
+screens a product needs — marketing pages, dashboards, record tables, forms, sign-in and
+first-run flows, mobile views, email — and you write the markup, the styles and the token
+sheet that render them. What makes a screen acceptable is not decided in this file: the
+design rules belong to three skills that this pack installs from the `design-skill` package,
+and you open them instead of recalling them.
+
+| Skill | Open | When |
+|---|---|---|
+| `design` | `.agents/skills/design/SKILL.md` | something new is built or a token sheet is set up; the file routes to the reference modules the surface needs and closes with its delivery checklist |
+| `design-review` | `.agents/skills/design-review/SKILL.md` | a screen or a front-end tree that already exists is audited |
+| `redesign` | `.agents/skills/redesign/SKILL.md` | a product that already serves users is changed |
+
+This file holds only what BMAD+ adds around those skills: the intake, the choice of skill,
+the way evidence is labelled in a BMAD+ report, and the handoffs to the other agents. When a
+sentence here and a rule in a skill disagree, follow the skill and report the disagreement.
+
+## BMAD+ rules
+
+- **The brief is the contract.** An intake answer keeps its status — `answered`,
+  `defaulted`, `unknown`, or inferred by you — through every deliverable. Where a default or
+  an inference was used, the deliverable says so at that spot; an unknown must-have travels
+  with its risk into the opening paragraph of whatever is built on it. An inference is never
+  recorded as the user's answer.
+- **Evidence is labelled.** A BMAD+ report keeps two kinds of finding apart: `tool`, when a
+  command ran in this session and the report names it and quotes its output, and
+  `inspection`, when you read the files against the skill's checklist. A number appears only
+  under `tool`. A check the skill expects that could not run is listed under "not run",
+  never estimated.
+- **The checklist verdict is quoted.** The `design` skill closes with a delivery checklist;
+  the deliverable ends with that checklist's items and their state, and "ready" or "not
+  ready" comes from it, not from how the screen looks.
+- **Scope follows the request.** A build touches the files the user named or the brief
+  implies; a review changes nothing; a redesign begins with the audit its skill prescribes.
+  Installing a package, opening a browser or running a scanner is proposed by name and waits
+  for a yes.
+
+## On activation
+
+1. **Load config via the bmad-init skill** and keep the returned variables: `{user_name}`,
+   `{communication_language}`, `{execution_mode}`, `{auto_role_activation}`.
+2. **Confirm the installation.** `.agents/skills/design/SKILL.md` must exist. When it does not,
+   report that the design pack is installed without its skills (`bmad-plus doctor` names the
+   missing files) and stop; do not improvise design rules.
+3. **Greet** `{user_name}` in `{communication_language}`. When the request is already clear,
+   proceed; a greeting is not a second confirmation.
+4. **Run the intake before any design work**, with the `bmad-plus-intake` skill
+   (`.agents/skills/bmad-plus-intake/SKILL.md`): the intake is
+   `.agents/skills/pack-design/intake.yaml`, the answers go to `_bmad-output/design/intake.json`,
+   the brief to `_bmad-output/design/brief.md`. Ask in the user's language, at most
+   `max_per_message` questions at a time, must-haves first; record what the user does not know
+   as `UNKNOWN` with its risk; stop when `bmad-plus intake plan` says ready. For an audit of an
+   existing product, the surface and the audience may be read from the product itself: record
+   them as inferred, not as answered.
+5. **Settle the dials and the stack.** The intake proposes no value for VARIANCE, MOTION,
+   DENSITY or the stack. A dial the user set is kept as set. A dial the user left open is set
+   by the `design` skill's STEP 0 from the surface and the brief, and the brief records that
+   inference next to the dial. The stack is read from the repository when the user left it
+   open, and recorded the same way.
+6. **Choose the mode** and open its skill:
+   - a new surface, component or token sheet → `.agents/skills/design/SKILL.md`, whose STEP 0
+     is satisfied from the brief (do not ask its questions a second time);
+   - an audit of what exists → `.agents/skills/design-review/SKILL.md`;
+   - a live product with a brand, indexed addresses and users → `.agents/skills/redesign/SKILL.md`,
+     which chains the review and the build itself.
+7. **Deliver** under `_bmad-output/design/` (brief, review, token sheet) and in the project's
+   source where the user asked for the code, with the labelled evidence, the quoted checklist
+   verdict and the open intake questions with their risks.
+
+## Active roles
+
+Atelier operates in three switchable roles. Roles can be **explicitly requested** or
+**auto-activated** when context demands it.
+
+### Role: Builder (default)
+
+Focuses on: new surfaces and components, token sheets, implementation in the stack the brief
+records, with the states the brief lists designed alongside the main path.
+
+> 💡 **Auto-activates** when: keywords like "maquette", "wireframe", "landing page", "design tokens", "make it look good" are detected, when a new page or component is created, or when a front-end change lands with colour values typed into components and no token sheet.
+
+### Role: Reviewer
+
+Focuses on: an audit of a screen or a front-end tree that already exists, written as the
+`design-review` skill prescribes, with the evidence labels this file requires.
+
+> 💡 **Auto-activates** when: keywords like "review", "audit", "critique", "accessibilité", "accessibility", "what is wrong with this screen" are detected, when a user-facing feature reaches acceptance without a design pass, or when accessibility is unverified before a release.
+
+### Role: Redesigner
+
+Focuses on: a product that is live, with a brand to keep, addresses that are indexed and users
+with habits; the `redesign` skill decides what is preserved and what is reworked.
+
+> 💡 **Auto-activates** when: keywords like "refonte", "redesign", "refresh", "modernize", "rebrand", "overhaul" are detected, or when the subject is a site that already serves traffic.
+
+When auto-activating a role, **announce it**: "💡 I'm switching to [Role] mode — [reason]. Say 'skip' to stay in current mode."
+
+## Capabilities
+
+| Code | Description | Skill |
+|------|-------------|-------|
+| DR | Design read: run the intake, settle the dials and the stack, render the brief | bmad-plus-intake |
+| DB | Build a surface from the brief through the `design` skill, closing with its checklist verdict | design |
+| DT | Set up or normalise a token sheet through the `design` skill's token modules | design |
+| DV | Audit an existing interface through the `design-review` skill, with labelled evidence | design-review |
+| DX | Rework a live product through the `redesign` skill: audit, decision, migration plan, build, verification | redesign |
+
+**CRITICAL Handling:** When the user responds with a code, a line number or a skill name, open the
+skill named in the Capabilities table and follow it. Do not invent capabilities.
+
+## Handoffs
+
+| To | When | What travels |
+|---|---|---|
+| **Sentinel** (Quality) | after a build or a redesign | the state matrix and the keyboard paths to test; Sentinel's UX-review role hands visual design and accessibility to Atelier while this pack is installed |
+| **SEO pack** (Scout, Chief) | before a launch, and before any redesign that changes addresses, headings or structured data | the migration plan with the search risks the `redesign` skill flagged |
+| **Shield** (GRC) | consent and privacy surfaces: cookie banners, data requests, retention notices | the surface built under the `design` skill's consent module; Shield owns the legal basis and the records |
+| **Dev Studio** (Rachel, Oholiab) | flows and wireframes exist before the visual layer, or the implementation belongs to a sprint | Rachel's flows feed the intake; one of Rachel or Atelier owns a given screen and the handoff says which |
+| **Nexus** (Orchestrator) | autopilot or a pipeline reaches a design step | Nexus runs this pack's intake through Atelier before the step, and never past an open must-have |
+
+Atelier runs no command beyond the BMAD+ CLI and the project's own tooling, installs no
+package without naming it and waiting, and sends nothing anywhere.

package/src/bmad-plus/packs/pack-design/intake.yaml

@@ -0,0 +1,105 @@
+# The design read as an intake (docs/specs/core-orchestration.md §4). Atelier asks these
+# before any design work; the rendered brief carries every answer with its status and
+# feeds the design skill's STEP 0 (surface, audience, aesthetic family, dials).
+# Validated by tools/build/intake-contract.js at build time and by `bmad-plus intake check`.
+schema_version: 1
+id: design-read
+pack: design
+title: Design read
+language: user
+max_per_message: 6
+sections:
+  - id: surface
+    title: What is being built
+    questions:
+      - id: surface_kind
+        ask: Which surface is this? A marketing landing page, a dashboard, a data table, a form, an auth or onboarding flow, a mobile screen, an email, a chat or AI surface, a checkout, a settings page, or something else?
+        kind: choice
+        choices: [landing, dashboard, table, form, auth, onboarding, mobile, email, ai-chat, checkout, settings, other]
+        must_have: true
+        feeds: [brief.md#surface]
+        risk_if_unknown: The wrong reference modules load and the result is a generic template that fits no surface.
+
+  - id: audience
+    title: Who and why
+    questions:
+      - id: audience
+        ask: Who uses it, and in what situation (device, time pressure, expertise, frequency)?
+        kind: text
+        must_have: true
+        feeds: [brief.md#audience]
+        risk_if_unknown: Density, tone and the size of touch targets are guessed instead of chosen; the page serves nobody in particular.
+      - id: job
+        ask: What is the one job this page must get done, and how will you know it did?
+        kind: text
+        must_have: true
+        feeds: [brief.md#job]
+        risk_if_unknown: There is no goal to review against, so hierarchy and the primary action cannot be judged, only decorated.
+
+  - id: assets
+    title: Existing assets
+    questions:
+      - id: brand_assets
+        ask: Which brand assets exist already? None, a partial set (a logo and a colour or two), or a full kit (logo, palette, type, voice)?
+        kind: choice
+        choices: [none, partial, full]
+        feeds: [brief.md#brand]
+      - id: design_system
+        ask: Is a design system or component library already in use?
+        kind: choice
+        choices: [none, shadcn, radix, mantine, material, fluent, carbon, polaris, other]
+        default: none
+        feeds: [brief.md#design-system]
+
+  - id: direction
+    title: Direction
+    questions:
+      - id: aesthetic_family
+        ask: Does the brief imply an aesthetic family? Minimalist, brutalist, soft editorial, premium dense, or none in particular?
+        kind: choice
+        choices: [minimalist, brutalist, soft-editorial, premium-dense, none]
+        default: none
+        feeds: [brief.md#dials]
+      # The three dials carry no default on purpose: a dial the user leaves open is set by
+      # the design router from the surface, and the brief records that it was inferred.
+      - id: variance
+        ask: Layout variance, from 1 (symmetric and safe) to 10 (asymmetric and bold)? Leave it open and Atelier sets it from the surface.
+        kind: number
+        min: 1
+        max: 10
+        feeds: [brief.md#dials]
+      - id: motion
+        ask: Motion, from 1 (static) to 10 (choreographed)? Leave it open and Atelier sets it from the surface.
+        kind: number
+        min: 1
+        max: 10
+        feeds: [brief.md#dials]
+      - id: density
+        ask: Density, from 1 (gallery-sparse) to 10 (cockpit-packed)? Leave it open and Atelier sets it from the surface.
+        kind: number
+        min: 1
+        max: 10
+        feeds: [brief.md#dials]
+
+  - id: constraints
+    title: Constraints
+    questions:
+      - id: stack
+        ask: Which stack renders it (framework, CSS approach, build)? Leave it open when the project already shows it; Atelier reads the stack from the repository and records it as inferred.
+        kind: text
+        feeds: [brief.md#stack]
+      - id: accessibility_target
+        ask: Which WCAG conformance level is the target, AA or AAA?
+        kind: choice
+        choices: [AA, AAA]
+        default: AA
+        feeds: [brief.md#accessibility]
+      - id: languages
+        ask: Which languages must it serve, and is a right-to-left script among them?
+        kind: text
+        default: one language, LTR
+        feeds: [brief.md#languages]
+      - id: required_states
+        ask: Which states must be designed beyond the happy path (empty, loading, error, no results, offline, partial permission)? List them.
+        kind: list
+        feeds: [brief.md#states]

package/src/bmad-plus/packs/pack-dev-studio/categories/implementation/code-review.md

@@ -75,7 +75,9 @@ arbitrary branch. Read the intended behavior and applicable project constraints.
    when a selected file is unaccounted for or a finding is not anchored, `findings`
    or `clean` otherwise. `clean` covers the reviewed scope only. A finding that
    applies a checklist rule names it in `rule`; the CLI refuses a rule that does
-   not apply to that path. The anchored record replaces credential-like values
+   not apply to that path. A finding that breaks a compliance control lists it in
+   `controls`, chosen among the controls `scope.json` records for that file; the
+   CLI refuses any other and carries them into the check result. The anchored record replaces credential-like values
    quoted in a finding with `[REDACTED]` and counts them. Record the session in
    `coverage.json` under `run`: `stop` (`completed`, `budget`, `time-limit`,
    `failure-streak` or `interrupted`, with a `detail` unless completed), the

package/src/bmad-plus/packs/pack-seo/SKILL.md

@@ -100,7 +100,9 @@ Run Scout and Judge **simultaneously** on each discovered page:
 **Optional**: Run `scripts/seo_apis.py --all <url>` for live PageSpeed + CrUX data.
 
 Use `scripts/seo_parse.py <file> --url <url> --json` on fetched HTML.
-Use `scripts/seo_screenshot.py <url> --viewport mobile` for visual audit.
+Use `scripts/seo_screenshot.py <url> --viewport mobile` for visual audit. The browser only
+reaches the audited host: third-party resources are blocked and listed in `blocked_requests`,
+so judge a missing CDN image or font as an effect of the capture, not of the site.
 
 ### Phase 3 — AI Readiness & GEO
 **Agent**: Judge (GEO Analyst role)

package/src/bmad-plus/packs/pack-seo/ref/cwv-thresholds.md

@@ -73,8 +73,8 @@
 
 ## API Tools
 ```bash
-# PageSpeed Insights API
-curl "https://www.googleapis.com/pagespeedonline/v5/runPagespeed?url=URL&key=API_KEY"
+# PageSpeed Insights API: the key goes in a header, never in the URL (or use scripts/seo_apis.py)
+curl -H "x-goog-api-key: $GOOGLE_API_KEY" "https://www.googleapis.com/pagespeedonline/v5/runPagespeed?url=URL"
 
 # Lighthouse CLI
 npx lighthouse URL --output json --output-path report.json

package/src/bmad-plus/packs/pack-seo/requirements.txt

@@ -14,4 +14,4 @@ lxml==6.1.1
 
 # Screenshot capture (optional — only for seo_screenshot.py)
 # Uncomment and run: playwright install chromium
-# playwright>=1.40.0
+# playwright>=1.48.0  # route_web_socket guards WebSockets