wize-dev-kit 0.7.2 → 0.8.0

package/CHANGELOG.md

@@ -5,6 +5,28 @@ Format inspired by [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
 ## [Unreleased]
 
+## [0.8.0] — 2026-07-04
+
+### Added
+
+- **OpenCode: commands agora rodam sob a persona dona.** O adapter deriva `agent: <code>` do `owner:` de cada workflow/skill (código, nome de exibição ou combinações "X + Y"), então `/wize-dev-story` já ativa o system prompt da Shuri em vez de depender de qual agente estava ativo.
+- **OpenCode: fan-out isolado nativamente.** `wize-review-adversarial` e `wize-review-edge-case-hunter` — os dois workers nomeados que o `wize-code-review` dispara em paralelo — ganham `subtask: true`, isolando-os do contexto de quem invoca independentemente do chamador.
+- **Escada de reuso (YAGNI) na persona da Shuri.** Antes de escrever código novo: precisa existir? já está no repo? a stdlib resolve? é nativo da plataforma? uma dependência instalada resolve? é uma linha? Só então código novo — referenciada por `wize-dev-story` e `wize-quick-dev`.
+- **Padrão de fan-out de subagentes documentado na persona do Wizer**, generalizado a partir do `wize-code-review`: nomear cada worker, escopar contexto, despachar pela primitiva nativa do harness (Task/Agent tool no Claude Code; `mode`/`subtask` no OpenCode; geração de prompts + fallback manual nos demais), tolerar falha parcial.
+- **`docs/harnesses/`** — um doc por harness suportada (9 adapters, em inglês + pt-BR), com overview curto e tabela no README apontando pra cada um. `package.json` passa a publicar `docs/` no pacote npm.
+
+### Fixed
+
+- **Instalador não tenta mais abrir a harness detectada ao final do `install`.** Em vez do prompt "abrir agora com Wizer?", sugere o comando inicial certo pro estado do repo (greenfield/brownfield) via `composeOnboarding`.
+- Comentário desatualizado no adapter do Kimi Code, que ainda citava o path antigo do Codex (`.codex/skills/`).
+
+## [0.7.3] — 2026-06-27
+
+### Fixed
+
+- **Codex adapter alinhado com a documentação pública da OpenAI.** O target `codex` voltou a renderizar skills em `.agents/skills/`, que é o diretório documentado para skills locais de repositório no Codex. A regressão em `0.7.2` tinha movido a saída para `.codex/skills/`, o que quebrou o carregamento das skills no harness real.
+- **Doctor, smoke tests e `.gitignore` corrigidos para o path canônico do Codex.**
+
 ## [0.7.2] — 2026-06-27
 
 ### Fixed

package/README.md

@@ -66,6 +66,24 @@ After install, open your IDE and say:
 
 ---
 
+## Supported harnesses
+
+All 9 IDE targets render from the same source; format and mechanics differ per harness. **OpenCode** gets the deepest integration — the kit's persona/workflow split maps onto OpenCode's own primitives (`mode: primary|subagent`, `agent:`, `subtask:`) instead of being flattened into one file type.
+
+| Harness | Output | Notable |
+|---|---|---|
+| **OpenCode** 🆕 | `.opencode/agents/` + `.opencode/commands/` | Native `mode: primary\|subagent`; commands auto-bind to their owning persona (`agent:`); fan-out workers run isolated (`subtask: true`). [Docs →](docs/harnesses/opencode.md) |
+| **Claude Code** | `.claude/skills/*/SKILL.md` | Anthropic Skill format; ad-hoc Task/Agent-tool fan-out (`wize-code-review`). [Docs →](docs/harnesses/claude-code.md) |
+| **Codex** | `.agents/skills/*/SKILL.md` | Same Skill format + root `AGENTS.md`. [Docs →](docs/harnesses/codex.md) |
+| **Kimi Code** | `.kimi/skills/*/SKILL.md` | Same Skill format; auto-detects Claude/Codex skill trees. [Docs →](docs/harnesses/kimi-code.md) |
+| **Antigravity** | `.agent/skills/*/SKILL.md` | Same Skill format + root `AGENTS.md`. [Docs →](docs/harnesses/antigravity.md) |
+| **Cursor** | `.cursor/rules/*.mdc` | On-demand rules (`alwaysApply: false`), matched by description. [Docs →](docs/harnesses/cursor.md) |
+| **Windsurf** | `.windsurf/rules/*.md` | Plain markdown; activation mode set inside the IDE. [Docs →](docs/harnesses/windsurf.md) |
+| **Continue.dev** | `.continue/prompts/*.prompt` | `invokable: true` slash commands. [Docs →](docs/harnesses/continue.md) |
+| **Generic fallback** | `.wize/agents/*.md` + root `AGENTS.md` | For any IDE without a dedicated adapter. [Docs →](docs/harnesses/generic.md) |
+
+---
+
 ## The roster
 
 | # | Persona | Code | Role |
@@ -203,12 +221,13 @@ npx wize-dev-kit uninstall       # remove .wize/ (your code is left untouched)
 - [`ROSTER.md`](ROSTER.md) — personas with style, role, BMAD equivalences.
 - [`DECISIONS.md`](DECISIONS.md) — decisions log.
 - [`CHANGELOG.md`](CHANGELOG.md) — release history.
+- [`docs/harnesses/`](docs/harnesses/) — one doc per [supported harness](#supported-harnesses), in English + [pt-BR](README.pt-BR.md#harnesses-suportadas).
 
 ---
 
 ## Status
 
-**v0.7.0 — beta.** The full lifecycle (analysis → plan → solution → implementation) is wired with 10 agents and a structured skill library. The `security-overlay` (AI Pentester) ships a complete pentest pipeline, a stakeholder report (risk score + briefing + AI action plan), and post-scan remediation planning — validated end-to-end against a real Laravel/PHP app. IDE adapters for Claude Code, Cursor, Windsurf, Codex, Continue, Kimi Code, OpenCode, and Antigravity are regenerated automatically.
+**v0.8.0 — beta.** The full lifecycle (analysis → plan → solution → implementation) is wired with 10 agents and a structured skill library. The `security-overlay` (AI Pentester) ships a complete pentest pipeline, a stakeholder report (risk score + briefing + AI action plan), and post-scan remediation planning — validated end-to-end against a real Laravel/PHP app. IDE adapters for Claude Code, Cursor, Windsurf, Codex, Continue, Kimi Code, OpenCode, and Antigravity are regenerated automatically — [OpenCode](docs/harnesses/opencode.md) gets native `mode`/`agent`/`subtask` wiring, the deepest integration of the 9.
 
 ---
 

package/README.pt-BR.md

@@ -66,6 +66,24 @@ Após instalar, abra sua IDE e diga:
 
 ---
 
+## Harnesses suportadas
+
+Os 9 alvos de IDE são gerados a partir da mesma fonte; formato e mecânica mudam por harness. O **OpenCode** tem a integração mais profunda — a divisão persona/workflow do kit mapeia pras primitivas nativas do próprio OpenCode (`mode: primary|subagent`, `agent:`, `subtask:`) em vez de ser achatada num único tipo de arquivo.
+
+| Harness | Saída | Destaque |
+|---|---|---|
+| **OpenCode** 🆕 | `.opencode/agents/` + `.opencode/commands/` | `mode: primary\|subagent` nativo; commands se ligam à persona dona (`agent:`); workers de fan-out rodam isolados (`subtask: true`). [Docs →](docs/harnesses/opencode.pt-BR.md) |
+| **Claude Code** | `.claude/skills/*/SKILL.md` | Formato Skill da Anthropic; fan-out ad hoc via Task/Agent tool (`wize-code-review`). [Docs →](docs/harnesses/claude-code.pt-BR.md) |
+| **Codex** | `.agents/skills/*/SKILL.md` | Mesmo formato Skill + `AGENTS.md` na raiz. [Docs →](docs/harnesses/codex.pt-BR.md) |
+| **Kimi Code** | `.kimi/skills/*/SKILL.md` | Mesmo formato Skill; autodetecta as árvores do Claude/Codex. [Docs →](docs/harnesses/kimi-code.pt-BR.md) |
+| **Antigravity** | `.agent/skills/*/SKILL.md` | Mesmo formato Skill + `AGENTS.md` na raiz. [Docs →](docs/harnesses/antigravity.pt-BR.md) |
+| **Cursor** | `.cursor/rules/*.mdc` | Rules sob demanda (`alwaysApply: false`), casadas por descrição. [Docs →](docs/harnesses/cursor.pt-BR.md) |
+| **Windsurf** | `.windsurf/rules/*.md` | Markdown puro; modo de ativação definido dentro da IDE. [Docs →](docs/harnesses/windsurf.pt-BR.md) |
+| **Continue.dev** | `.continue/prompts/*.prompt` | Slash commands via `invokable: true`. [Docs →](docs/harnesses/continue.pt-BR.md) |
+| **Fallback genérico** | `.wize/agents/*.md` + `AGENTS.md` na raiz | Para qualquer IDE sem adapter dedicado. [Docs →](docs/harnesses/generic.pt-BR.md) |
+
+---
+
 ## O elenco
 
 | # | Persona | Código | Papel |
@@ -203,12 +221,13 @@ npx wize-dev-kit uninstall       # remove .wize/ (seu código permanece intacto)
 - [`ROSTER.md`](ROSTER.md) — personas com estilo, papel, equivalências BMAD.
 - [`DECISIONS.md`](DECISIONS.md) — log de decisões.
 - [`CHANGELOG.md`](CHANGELOG.md) — histórico de releases.
+- [`docs/harnesses/`](docs/harnesses/) — um doc por [harness suportada](#harnesses-suportadas), em português + [English](README.md#supported-harnesses).
 
 ---
 
 ## Status
 
-**v0.7.0 — beta.** O ciclo completo (análise → plano → solução → implementação) está montado com 10 agentes e uma biblioteca estruturada de skills. O `security-overlay` (Pentester de IA) entrega um pipeline de pentest completo, um relatório executivo (score de risco + briefing + plano de ação por IA) e planejamento de correção pós-scan — validado de ponta a ponta contra uma aplicação Laravel/PHP real. Os adapters de IDE para Claude Code, Cursor, Windsurf, Codex, Continue, Kimi Code, OpenCode e Antigravity são regenerados automaticamente.
+**v0.8.0 — beta.** O ciclo completo (análise → plano → solução → implementação) está montado com 10 agentes e uma biblioteca estruturada de skills. O `security-overlay` (Pentester de IA) entrega um pipeline de pentest completo, um relatório executivo (score de risco + briefing + plano de ação por IA) e planejamento de correção pós-scan — validado de ponta a ponta contra uma aplicação Laravel/PHP real. Os adapters de IDE para Claude Code, Cursor, Windsurf, Codex, Continue, Kimi Code, OpenCode e Antigravity são regenerados automaticamente — o [OpenCode](docs/harnesses/opencode.pt-BR.md) tem a integração mais profunda das 9, com `mode`/`agent`/`subtask` nativos.
 
 ---
 

package/adapters/README.md

@@ -9,7 +9,7 @@ Each adapter renders the kit's agents/skills/workflows into the file layout that
 | `claude-code` | `.claude/skills/wize-*/` | `SKILL.md` per agent/workflow | Default. Uses Claude Code skill folder pattern. |
 | `cursor` | `.cursor/rules/wize-*.mdc` | MDC | Each agent/skill becomes a rule file. |
 | `windsurf` | `.windsurf/agents/wize-*.md` | Markdown | Cascade-friendly. |
-| `codex` | `.codex/skills/wize-*/` | `SKILL.md` per agent/workflow | OpenAI Codex. |
+| `codex` | `.agents/skills/wize-*/` | `SKILL.md` per agent/workflow | OpenAI Codex. |
 | `continue` | `.continue/agents/wize-*.md` | Markdown | Continue agent slot. |
 | `kimi-code` | `.kimi/agents/wize-*.md` | Markdown | Moonshot Kimi Code. |
 | `opencode` | `.opencode/agents/wize-*.md` | Markdown | OpenCode CLI. |

package/adapters/codex/README.md

@@ -2,7 +2,7 @@
 
 OpenAI Codex skills.
 
-- **Target path:** `.codex/skills/`
+- **Target path:** `.agents/skills/`
 - **File pattern:** `wize-*/SKILL.md`
 - **Format:** Anthropic-compatible `SKILL.md`
 

package/adapters/codex/adapter.yaml

@@ -1,6 +1,6 @@
 code: codex
-description: "OpenAI Codex agent skills (.codex/skills/, Anthropic-compatible)."
-target_path: ".codex/skills/"
+description: "OpenAI Codex agent skills (.agents/skills/, Anthropic-compatible)."
+target_path: ".agents/skills/"
 file_pattern: "wize-*/SKILL.md"
 file_format: markdown
 hooks:

package/adapters/codex/render.js

@@ -1,4 +1,4 @@
-// OpenAI Codex adapter — emits .codex/skills/wize-{code}/SKILL.md per asset.
+// OpenAI Codex adapter — emits .agents/skills/wize-{code}/SKILL.md per asset.
 // Codex shares the Anthropic SKILL.md format and also reads the root AGENTS.md.
 
 'use strict';
@@ -7,7 +7,7 @@ const path = require('node:path');
 const { renderAnthropicSkills } = require('../../tools/installer/render-shared.js');
 
 function render(kitRoot, projectRoot, opts = {}) {
-  const targetDir = path.join(projectRoot, '.codex', 'skills');
+  const targetDir = path.join(projectRoot, '.agents', 'skills');
   return renderAnthropicSkills(kitRoot, targetDir, opts);
 }
 

package/adapters/kimi-code/render.js

@@ -1,7 +1,7 @@
 // Moonshot Kimi Code adapter — emits .kimi/skills/wize-{code}/SKILL.md per asset.
 // Kimi Code consumes the Anthropic SKILL.md format verbatim and additionally
-// auto-detects .claude/skills/ and .codex/skills/, so installing alongside
-// those adapters is harmless.
+// auto-detects .claude/skills/ and .agents/skills/ (Codex), so installing
+// alongside those adapters is harmless.
 
 'use strict';
 

package/adapters/opencode/render.js

@@ -2,7 +2,10 @@
 //   - .opencode/agents/wize-{code}.md     (personas → system prompts)
 //   - .opencode/commands/wize-{code}.md   (workflows/skills → slash commands)
 // Frontmatter (agents): description, mode (primary|subagent|all)
-// Frontmatter (commands): description, template (the prompt body)
+// Frontmatter (commands): description, agent (derived from the source's
+//   `owner:` field, when it resolves to a known persona), subtask (from the
+//   source's `subtask: true`, for skills meant to run isolated regardless of
+//   which agent invokes them — e.g. the wize-code-review fan-out workers)
 // Reference: https://opencode.ai/docs/agents/ + /docs/commands/
 
 'use strict';
@@ -11,10 +14,30 @@ const fs = require('node:fs');
 const path = require('node:path');
 const { collectAssets, escapeYamlDouble, clipOneLine, ensureDir } = require('../../tools/installer/render-shared.js');
 
+// `owner:` in workflow/skill frontmatter is written for humans (agent code,
+// display name, "X + Y" pairings, trailing "# comment"). Resolve it to a
+// single known agent code, or null if it doesn't clearly name one (e.g.
+// builder-skills' `owner: builder`, which isn't a persona).
+function resolveOwnerAgentCode(ownerRaw, agents) {
+  if (!ownerRaw) return null;
+  const withoutComment = ownerRaw.split('#')[0].trim();
+  const primary = withoutComment.split(/\s*\+\s*|,/)[0].trim();
+  if (!primary) return null;
+  const byCode = agents.find(a => a.code === primary);
+  if (byCode) return byCode.code;
+  const lower = primary.toLowerCase();
+  const byName = agents.find(a => {
+    const name = a.name.toLowerCase();
+    return name === lower || name.includes(lower) || lower.includes(name.split(' ')[0]);
+  });
+  return byName ? byName.code : null;
+}
+
 function render(kitRoot, projectRoot, opts = {}) {
   const agentsDir = path.join(projectRoot, '.opencode', 'agents');
   const commandsDir = path.join(projectRoot, '.opencode', 'commands');
   const assets = collectAssets(kitRoot, opts);
+  const agents = assets.filter(a => a.kind === 'agent');
   const written = [];
 
   if (!opts.dryRun) {
@@ -46,9 +69,12 @@ function render(kitRoot, projectRoot, opts = {}) {
     } else {
       // workflow or skill → command
       const desc = clipOneLine(a.description, 180);
+      const agentCode = resolveOwnerAgentCode(a.owner, agents);
       const content = [
         '---',
         `description: "${escapeYamlDouble(desc)}"`,
+        ...(agentCode ? [`agent: ${agentCode}`] : []),
+        ...(a.subtask ? ['subtask: true'] : []),
         '---',
         '',
         `# ${a.name}`,

package/docs/harnesses/antigravity.md

@@ -0,0 +1,20 @@
+# Google Antigravity — Wize Development Kit
+
+🌐 **Languages:** **English** · [Português (pt-BR)](antigravity.pt-BR.md)
+
+← [Back to README](../../README.md)
+
+Antigravity uses the same public Anthropic Skill format as Claude Code and Codex.
+
+## Output
+
+- `.agent/skills/wize-{code}/SKILL.md` — note the **singular** `.agent`, not `.agents` (that's Codex's directory).
+- A root `AGENTS.md` (shared with Codex, Cursor, and Windsurf).
+
+## Notable
+
+- `.antigravitycli/` is Antigravity's own CLI state directory and is never touched by the kit — only `.agent/skills/` and `AGENTS.md` are written.
+
+## Setup
+
+Pick **Antigravity** as an IDE target during `npx wize-dev-kit install` (or add it and re-run `npx wize-dev-kit sync`). Restart Antigravity, then run `/wize-orchestrator` or ask it to read `AGENTS.md`.

package/docs/harnesses/antigravity.pt-BR.md

@@ -0,0 +1,20 @@
+# Google Antigravity — Wize Development Kit
+
+🌐 **Idiomas:** [English](antigravity.md) · **Português (pt-BR)**
+
+← [Voltar ao README](../../README.pt-BR.md)
+
+O Antigravity usa o mesmo formato público de Skill da Anthropic que o Claude Code e o Codex.
+
+## Saída
+
+- `.agent/skills/wize-{code}/SKILL.md` — repare no **singular** `.agent`, não `.agents` (esse é o diretório do Codex).
+- Um `AGENTS.md` na raiz (compartilhado com Codex, Cursor e Windsurf).
+
+## Destaques
+
+- `.antigravitycli/` é o diretório de estado do próprio CLI do Antigravity e nunca é tocado pelo kit — só `.agent/skills/` e `AGENTS.md` são gravados.
+
+## Como ativar
+
+Escolha **Antigravity** como alvo de IDE em `npx wize-dev-kit install` (ou adicione depois e rode `npx wize-dev-kit sync`). Reinicie o Antigravity e rode `/wize-orchestrator` ou peça pra ele ler o `AGENTS.md`.

package/docs/harnesses/claude-code.md

@@ -0,0 +1,23 @@
+# Claude Code — Wize Development Kit
+
+🌐 **Languages:** **English** · [Português (pt-BR)](claude-code.pt-BR.md)
+
+← [Back to README](../../README.md)
+
+Claude Code was the kit's first target and uses the public Anthropic Skill format verbatim.
+
+## Output
+
+`.claude/skills/wize-{code}/SKILL.md` — one directory per asset (persona, workflow, or skill). YAML frontmatter (`name`, `description`) plus the Markdown body. Companion files a workflow references by relative path (`steps/`, `templates/`, `data/`, `customize.toml`, ...) are copied alongside the `SKILL.md` so those paths resolve.
+
+There is no separate "agent" file shape here — personas (Wizer, Shuri, Hawkeye...) and workflows/skills all render to the same `SKILL.md` structure.
+
+## Notable
+
+- **Discovery is description-based.** Claude Code surfaces installed skills by matching the task at hand against each `description`; you can also invoke one directly by name (`/wize-dev-story`) or ask to "activate Wizer."
+- **Subagent fan-out is ad hoc, not persona-bound.** `wize-code-review` uses Claude Code's own Task/Agent tool to spin up isolated, parallel workers (Blind Hunter, Edge Case Hunter, Acceptance Auditor) for a single review pass. Any skill can do this — it isn't wired to a specific `.claude/agents/*.md` subagent definition, since the kit doesn't emit one.
+- Claude Code does **not** read the root `AGENTS.md` the installer generates (that file is for Codex, Cursor, Windsurf, and Antigravity).
+
+## Setup
+
+Pick **Claude Code** as an IDE target during `npx wize-dev-kit install` (default). Restart Claude Code, then say *"Activate Wizer and brief him on the project"* or run `/wize-orchestrator`.

package/docs/harnesses/claude-code.pt-BR.md

@@ -0,0 +1,23 @@
+# Claude Code — Wize Development Kit
+
+🌐 **Idiomas:** [English](claude-code.md) · **Português (pt-BR)**
+
+← [Voltar ao README](../../README.pt-BR.md)
+
+Claude Code foi o primeiro alvo do kit e usa o formato público de Skill da Anthropic ao pé da letra.
+
+## Saída
+
+`.claude/skills/wize-{code}/SKILL.md` — um diretório por asset (persona, workflow ou skill). Frontmatter YAML (`name`, `description`) mais o corpo em Markdown. Arquivos complementares que um workflow referencia por caminho relativo (`steps/`, `templates/`, `data/`, `customize.toml`, ...) são copiados junto do `SKILL.md` para que esses caminhos resolvam.
+
+Não existe um formato separado de "agente" aqui — personas (Wizer, Shuri, Hawkeye...) e workflows/skills viram todos a mesma estrutura de `SKILL.md`.
+
+## Destaques
+
+- **Descoberta por descrição.** O Claude Code encontra as skills instaladas casando a tarefa em mãos contra cada `description`; também dá pra invocar direto pelo nome (`/wize-dev-story`) ou pedir pra "ativar o Wizer".
+- **Fan-out de subagentes é ad hoc, não amarrado a persona.** O `wize-code-review` usa a própria ferramenta Task/Agent do Claude Code pra levantar workers isolados e paralelos (Blind Hunter, Edge Case Hunter, Acceptance Auditor) numa única passada de review. Qualquer skill pode fazer isso — não está amarrado a uma definição de subagente `.claude/agents/*.md`, já que o kit não gera nenhuma.
+- O Claude Code **não** lê o `AGENTS.md` na raiz que o instalador gera (esse arquivo é pro Codex, Cursor, Windsurf e Antigravity).
+
+## Como ativar
+
+Escolha **Claude Code** como alvo de IDE em `npx wize-dev-kit install` (padrão). Reinicie o Claude Code e diga *"Ative o Wizer e dê o briefing do projeto a ele"* ou rode `/wize-orchestrator`.

package/docs/harnesses/codex.md

@@ -0,0 +1,21 @@
+# OpenAI Codex — Wize Development Kit
+
+🌐 **Languages:** **English** · [Português (pt-BR)](codex.pt-BR.md)
+
+← [Back to README](../../README.md)
+
+Codex shares the same Anthropic Skill format as Claude Code and Kimi Code.
+
+## Output
+
+- `.agents/skills/wize-{code}/SKILL.md` — one directory per persona, workflow, or skill.
+- A root `AGENTS.md` (shared with Cursor, Windsurf, and Antigravity) listing the roster and pointing at `wize-orchestrator` as the entry point.
+
+## Notable
+
+- `.agents/skills/` is Codex's documented path for repository-local skills — the kit writes there directly, not to a Codex-specific directory.
+- **History:** 0.7.2 briefly moved this output to `.codex/skills/`, which broke skill loading in real Codex sessions; 0.7.3 reverted it to `.agents/skills/`. If you're on an older install, run `npx wize-dev-kit update` to fix the path.
+
+## Setup
+
+Pick **Codex** as an IDE target during `npx wize-dev-kit install` (or add it and re-run `npx wize-dev-kit sync`). Restart Codex, then run `/wize-orchestrator` or ask it to read `AGENTS.md`.

package/docs/harnesses/codex.pt-BR.md

@@ -0,0 +1,21 @@
+# OpenAI Codex — Wize Development Kit
+
+🌐 **Idiomas:** [English](codex.md) · **Português (pt-BR)**
+
+← [Voltar ao README](../../README.pt-BR.md)
+
+O Codex usa o mesmo formato de Skill da Anthropic que o Claude Code e o Kimi Code.
+
+## Saída
+
+- `.agents/skills/wize-{code}/SKILL.md` — um diretório por persona, workflow ou skill.
+- Um `AGENTS.md` na raiz (compartilhado com Cursor, Windsurf e Antigravity) listando o elenco e apontando pro `wize-orchestrator` como ponto de entrada.
+
+## Destaques
+
+- `.agents/skills/` é o caminho documentado do Codex pra skills locais do repositório — o kit grava lá diretamente, não num diretório específico do Codex.
+- **Histórico:** a 0.7.2 moveu essa saída, por um breve período, para `.codex/skills/`, o que quebrou o carregamento das skills em sessões reais do Codex; a 0.7.3 reverteu para `.agents/skills/`. Se sua instalação é antiga, rode `npx wize-dev-kit update` pra corrigir o caminho.
+
+## Como ativar
+
+Escolha **Codex** como alvo de IDE em `npx wize-dev-kit install` (ou adicione depois e rode `npx wize-dev-kit sync`). Reinicie o Codex e rode `/wize-orchestrator` ou peça pra ele ler o `AGENTS.md`.

package/docs/harnesses/continue.md

@@ -0,0 +1,21 @@
+# Continue.dev — Wize Development Kit
+
+🌐 **Languages:** **English** · [Português (pt-BR)](continue.pt-BR.md)
+
+← [Back to README](../../README.md)
+
+## Output
+
+`.continue/prompts/wize-{code}.prompt` — one file per persona, workflow, or skill.
+
+## Notable
+
+- Frontmatter: `name`, `description`, `invokable: true`. The `invokable` flag is what registers the file as a slash command in Continue's palette instead of a passive context document.
+
+## Setup
+
+Pick **Continue** as an IDE target during `npx wize-dev-kit install` (or add it and re-run `npx wize-dev-kit sync`). Reload the Continue extension, then run `/wize-orchestrator` from the prompt palette.
+
+## Reference
+
+- [docs.continue.dev — Prompts](https://docs.continue.dev/customize/deep-dives/prompts)

package/docs/harnesses/continue.pt-BR.md

@@ -0,0 +1,21 @@
+# Continue.dev — Wize Development Kit
+
+🌐 **Idiomas:** [English](continue.md) · **Português (pt-BR)**
+
+← [Voltar ao README](../../README.pt-BR.md)
+
+## Saída
+
+`.continue/prompts/wize-{code}.prompt` — um arquivo por persona, workflow ou skill.
+
+## Destaques
+
+- Frontmatter: `name`, `description`, `invokable: true`. É a flag `invokable` que registra o arquivo como slash command na paleta do Continue, em vez de um documento de contexto passivo.
+
+## Como ativar
+
+Escolha **Continue** como alvo de IDE em `npx wize-dev-kit install` (ou adicione depois e rode `npx wize-dev-kit sync`). Recarregue a extensão do Continue e rode `/wize-orchestrator` na paleta de prompts.
+
+## Referência
+
+- [docs.continue.dev — Prompts](https://docs.continue.dev/customize/deep-dives/prompts)

package/docs/harnesses/cursor.md

@@ -0,0 +1,22 @@
+# Cursor — Wize Development Kit
+
+🌐 **Languages:** **English** · [Português (pt-BR)](cursor.pt-BR.md)
+
+← [Back to README](../../README.md)
+
+## Output
+
+`.cursor/rules/wize-{code}.mdc` — one rule per persona, workflow, or skill.
+
+## Notable
+
+- Frontmatter: `description`, `globs` (empty — not file-pattern-triggered), `alwaysApply: false`. Every wize-dev-kit rule attaches **on demand**, matched by Cursor against its `description`, rather than being injected into every prompt.
+- No native agent/subagent split in this format — personas and workflows/skills share the same `.mdc` shape (unlike [OpenCode](opencode.md)).
+
+## Setup
+
+Pick **Cursor** as an IDE target during `npx wize-dev-kit install` (or add it and re-run `npx wize-dev-kit sync`). Restart Cursor, then reference `@wize-orchestrator` or ask it to activate Wizer.
+
+## Reference
+
+- [cursor.com/docs/context/rules](https://cursor.com/docs/context/rules)

package/docs/harnesses/cursor.pt-BR.md

@@ -0,0 +1,22 @@
+# Cursor — Wize Development Kit
+
+🌐 **Idiomas:** [English](cursor.md) · **Português (pt-BR)**
+
+← [Voltar ao README](../../README.pt-BR.md)
+
+## Saída
+
+`.cursor/rules/wize-{code}.mdc` — uma rule por persona, workflow ou skill.
+
+## Destaques
+
+- Frontmatter: `description`, `globs` (vazio — não é disparado por padrão de arquivo), `alwaysApply: false`. Toda rule do wize-dev-kit se anexa **sob demanda**, casada pelo Cursor contra sua `description`, em vez de ser injetada em todo prompt.
+- Sem divisão nativa agente/subagente nesse formato — personas e workflows/skills compartilham o mesmo formato `.mdc` (diferente do [OpenCode](opencode.pt-BR.md)).
+
+## Como ativar
+
+Escolha **Cursor** como alvo de IDE em `npx wize-dev-kit install` (ou adicione depois e rode `npx wize-dev-kit sync`). Reinicie o Cursor e referencie `@wize-orchestrator` ou peça pra ativar o Wizer.
+
+## Referência
+
+- [cursor.com/docs/context/rules](https://cursor.com/docs/context/rules)

package/docs/harnesses/generic.md

@@ -0,0 +1,20 @@
+# Generic fallback — Wize Development Kit
+
+🌐 **Languages:** **English** · [Português (pt-BR)](generic.pt-BR.md)
+
+← [Back to README](../../README.md)
+
+For any AI IDE that isn't one of the [dedicated targets](../../README.md#supported-harnesses).
+
+## Output
+
+- `.wize/agents/wize-{code}.md` — plain Markdown per persona, workflow, or skill: a heading, a one-line summary blockquote, and the body. No frontmatter, no slash-command wiring.
+- A root `AGENTS.md` — the same baseline pointer file also read natively by Codex, Cursor, Windsurf, and Antigravity. It is never overwritten if one already exists in the repo.
+
+## Notable
+
+- This target has no activation mechanism of its own — you point your assistant at `.wize/agents/wize-orchestrator.md` (or `AGENTS.md`) and ask it to read and follow it.
+
+## Setup
+
+Pick **Generic fallback** as an IDE target during `npx wize-dev-kit install` — it's on by default alongside your primary target, precisely so an unsupported second tool in the same repo still has something to read.

package/docs/harnesses/generic.pt-BR.md

@@ -0,0 +1,20 @@
+# Fallback genérico — Wize Development Kit
+
+🌐 **Idiomas:** [English](generic.md) · **Português (pt-BR)**
+
+← [Voltar ao README](../../README.pt-BR.md)
+
+Para qualquer IDE com IA que não esteja entre os [alvos dedicados](../../README.pt-BR.md#harnesses-suportadas).
+
+## Saída
+
+- `.wize/agents/wize-{code}.md` — Markdown puro por persona, workflow ou skill: um título, um resumo de uma linha em bloco de citação, e o corpo. Sem frontmatter, sem integração de slash command.
+- Um `AGENTS.md` na raiz — o mesmo arquivo-ponteiro de baseline lido nativamente por Codex, Cursor, Windsurf e Antigravity. Nunca é sobrescrito se já existir um no repositório.
+
+## Destaques
+
+- Esse alvo não tem mecanismo de ativação próprio — você aponta seu assistente pra `.wize/agents/wize-orchestrator.md` (ou `AGENTS.md`) e pede pra ele ler e seguir.
+
+## Como ativar
+
+Escolha **Fallback genérico** como alvo de IDE em `npx wize-dev-kit install` — ele já vem ligado por padrão junto do seu alvo principal, justamente pra que uma segunda ferramenta sem suporte dedicado no mesmo repositório ainda tenha algo pra ler.

package/docs/harnesses/kimi-code.md

@@ -0,0 +1,19 @@
+# Kimi Code — Wize Development Kit
+
+🌐 **Languages:** **English** · [Português (pt-BR)](kimi-code.pt-BR.md)
+
+← [Back to README](../../README.md)
+
+Moonshot's Kimi Code consumes the same public Anthropic Skill format as Claude Code and Codex.
+
+## Output
+
+`.kimi/skills/wize-{code}/SKILL.md` — one directory per persona, workflow, or skill, with companion files (`steps/`, `templates/`, `data/`) copied alongside.
+
+## Notable
+
+- Kimi Code also auto-detects `.claude/skills/` and `.agents/skills/` (Codex) if present in the repo, so installing the **Kimi Code** target alongside **Claude Code** and/or **Codex** is harmless — Kimi just has more places to find the same skills.
+
+## Setup
+
+Pick **Kimi Code** as an IDE target during `npx wize-dev-kit install` (or add it and re-run `npx wize-dev-kit sync`). Restart Kimi Code, then run `/wize-orchestrator`.

package/docs/harnesses/kimi-code.pt-BR.md

@@ -0,0 +1,19 @@
+# Kimi Code — Wize Development Kit
+
+🌐 **Idiomas:** [English](kimi-code.md) · **Português (pt-BR)**
+
+← [Voltar ao README](../../README.pt-BR.md)
+
+O Kimi Code, da Moonshot, consome o mesmo formato público de Skill da Anthropic usado pelo Claude Code e pelo Codex.
+
+## Saída
+
+`.kimi/skills/wize-{code}/SKILL.md` — um diretório por persona, workflow ou skill, com arquivos complementares (`steps/`, `templates/`, `data/`) copiados junto.
+
+## Destaques
+
+- O Kimi Code também autodetecta `.claude/skills/` e `.agents/skills/` (Codex) se existirem no repositório, então instalar o alvo **Kimi Code** junto com **Claude Code** e/ou **Codex** é inofensivo — o Kimi só passa a ter mais lugares onde achar as mesmas skills.
+
+## Como ativar
+
+Escolha **Kimi Code** como alvo de IDE em `npx wize-dev-kit install` (ou adicione depois e rode `npx wize-dev-kit sync`). Reinicie o Kimi Code e rode `/wize-orchestrator`.

package/docs/harnesses/opencode.md

@@ -0,0 +1,31 @@
+# OpenCode — Wize Development Kit
+
+🌐 **Languages:** **English** · [Português (pt-BR)](opencode.pt-BR.md)
+
+← [Back to README](../../README.md)
+
+OpenCode is the harness where the kit's persona/workflow split maps onto a **native** primitive instead of being flattened to one file type. Everywhere else, "agent" and "workflow" both become the same kind of file; on OpenCode they become two different things OpenCode itself understands.
+
+## Output
+
+`npx wize-dev-kit sync` (or `install`) writes two trees:
+
+- `.opencode/agents/wize-{code}.md` — the 10 personas (Wizer, Pepper, Shuri, Hawkeye...).
+- `.opencode/commands/wize-{code}.md` — the workflows and skills (`/wize-dev-story`, `/wize-code-review`, ...).
+
+## Notable
+
+- **`mode: primary | subagent`.** Only `wize-orchestrator` (Wizer) is `primary`. Every other persona is `subagent` — addressable directly with `@wize-agent-dev`, or delegated to automatically based on its `description`.
+- **`agent:` on commands.** Each workflow/skill's `owner:` frontmatter (e.g. `owner: wize-agent-dev # Shuri`) is resolved against the installed personas, so `/wize-dev-story` runs under Shuri's system prompt instead of whatever agent happened to be active. Commands with no clear owner (the Agent Builder skills) are left unbound.
+- **`subtask: true` on fan-out workers.** `wize-review-adversarial` and `wize-review-edge-case-hunter` — the two named subagent workers `wize-code-review` dispatches in parallel — are marked `subtask: true`, so OpenCode isolates them from the caller's context regardless of who invokes them.
+
+This is also the harness the kit's own subagent fan-out pattern (documented in Wizer's persona) is written against: name each worker, scope its context, dispatch on `mode: subagent` / `subtask: true`, tolerate partial failure.
+
+## Setup
+
+Pick **OpenCode** as an IDE target during `npx wize-dev-kit install` (or add it and re-run `npx wize-dev-kit sync`). Restart OpenCode, then say *"Activate Wizer and brief him on the project"* or run `/wize-orchestrator`.
+
+## Reference
+
+- [opencode.ai/docs/agents](https://opencode.ai/docs/agents/)
+- [opencode.ai/docs/commands](https://opencode.ai/docs/commands/)

package/docs/harnesses/opencode.pt-BR.md

@@ -0,0 +1,31 @@
+# OpenCode — Wize Development Kit
+
+🌐 **Idiomas:** [English](opencode.md) · **Português (pt-BR)**
+
+← [Voltar ao README](../../README.pt-BR.md)
+
+OpenCode é a harness onde a divisão persona/workflow do kit mapeia para uma primitiva **nativa**, em vez de ser achatada num único tipo de arquivo. Em todo o resto, "agente" e "workflow" viram o mesmo tipo de arquivo; no OpenCode eles viram duas coisas diferentes que o próprio OpenCode entende.
+
+## Saída
+
+`npx wize-dev-kit sync` (ou `install`) grava duas árvores:
+
+- `.opencode/agents/wize-{code}.md` — as 10 personas (Wizer, Pepper, Shuri, Hawkeye...).
+- `.opencode/commands/wize-{code}.md` — os workflows e skills (`/wize-dev-story`, `/wize-code-review`, ...).
+
+## Destaques
+
+- **`mode: primary | subagent`.** Só o `wize-orchestrator` (Wizer) é `primary`. Toda outra persona é `subagent` — endereçável direto com `@wize-agent-dev`, ou delegada automaticamente com base na `description`.
+- **`agent:` nos commands.** O campo `owner:` de cada workflow/skill (ex.: `owner: wize-agent-dev # Shuri`) é resolvido contra as personas instaladas, então `/wize-dev-story` roda sob o system prompt da Shuri em vez do agente que estava ativo. Commands sem um dono claro (as skills do Agent Builder) ficam sem vínculo.
+- **`subtask: true` nos workers de fan-out.** `wize-review-adversarial` e `wize-review-edge-case-hunter` — os dois workers nomeados que o `wize-code-review` dispara em paralelo — vêm marcados como `subtask: true`, então o OpenCode os isola do contexto de quem invocou, não importa quem seja.
+
+Esta também é a harness contra a qual o padrão de fan-out de subagentes do kit (documentado na persona da Wizer) foi escrito: nomear cada worker, escopar seu contexto, despachar via `mode: subagent` / `subtask: true`, tolerar falha parcial.
+
+## Como ativar
+
+Escolha **OpenCode** como alvo de IDE em `npx wize-dev-kit install` (ou adicione depois e rode `npx wize-dev-kit sync`). Reinicie o OpenCode e diga *"Ative o Wizer e dê o briefing do projeto a ele"* ou rode `/wize-orchestrator`.
+
+## Referência
+
+- [opencode.ai/docs/agents](https://opencode.ai/docs/agents/)
+- [opencode.ai/docs/commands](https://opencode.ai/docs/commands/)

package/docs/harnesses/windsurf.md

@@ -0,0 +1,22 @@
+# Windsurf (Codeium) — Wize Development Kit
+
+🌐 **Languages:** **English** · [Português (pt-BR)](windsurf.pt-BR.md)
+
+← [Back to README](../../README.md)
+
+## Output
+
+`.windsurf/rules/wize-{code}.md` — plain Markdown, no frontmatter: a heading, a one-line summary blockquote, and the full body.
+
+## Notable
+
+- Windsurf's Cascade reads everything under `.windsurf/rules/` at session start. Activation mode (always-on, manual, model-decided, glob) is configured **inside the IDE's Rules panel** per file — the kit doesn't encode it, since Windsurf's rule format has no frontmatter for it.
+- Same flattening as Cursor: no agent/workflow distinction in the file format.
+
+## Setup
+
+Pick **Windsurf** as an IDE target during `npx wize-dev-kit install` (or add it and re-run `npx wize-dev-kit sync`). Restart Windsurf, review the new rules under **Cascade → Rules**, then ask it to activate Wizer.
+
+## Reference
+
+- [docs.windsurf.com — Cascade rules](https://docs.windsurf.com/windsurf/cascade/memories#rules)

package/docs/harnesses/windsurf.pt-BR.md

@@ -0,0 +1,22 @@
+# Windsurf (Codeium) — Wize Development Kit
+
+🌐 **Idiomas:** [English](windsurf.md) · **Português (pt-BR)**
+
+← [Voltar ao README](../../README.pt-BR.md)
+
+## Saída
+
+`.windsurf/rules/wize-{code}.md` — Markdown puro, sem frontmatter: um título, um resumo de uma linha em bloco de citação, e o corpo completo.
+
+## Destaques
+
+- O Cascade do Windsurf lê tudo dentro de `.windsurf/rules/` no início da sessão. O modo de ativação (sempre ativo, manual, decidido pelo modelo, por glob) é configurado **dentro do painel de Rules da IDE**, por arquivo — o kit não codifica isso, já que o formato de rule do Windsurf não tem frontmatter pra isso.
+- Mesmo achatamento do Cursor: sem distinção agente/workflow no formato do arquivo.
+
+## Como ativar
+
+Escolha **Windsurf** como alvo de IDE em `npx wize-dev-kit install` (ou adicione depois e rode `npx wize-dev-kit sync`). Reinicie o Windsurf, confira as novas rules em **Cascade → Rules**, e peça pra ativar o Wizer.
+
+## Referência
+
+- [docs.windsurf.com — Cascade rules](https://docs.windsurf.com/windsurf/cascade/memories#rules)

package/package.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://json.schemastore.org/package.json",
   "name": "wize-dev-kit",
-  "version": "0.7.2",
+  "version": "0.8.0",
   "description": "Full-lifecycle AI-assisted development kit with Test Architect and Whiteport Design Studio embedded. Inspired by BMAD Method and WDS.",
   "keywords": [
     "ai",
@@ -42,6 +42,7 @@
     "adapters/",
     "tools/",
     "schemas/",
+    "docs/",
     "README.md",
     "README.pt-BR.md",
     "README.es.md",

package/src/core-skills/wize-review-adversarial/skill.md

@@ -3,6 +3,7 @@ code: wize-review-adversarial
 name: Adversarial Review
 module: core
 status: stub
+subtask: true
 ---
 
 # Adversarial Review

package/src/core-skills/wize-review-edge-case-hunter/skill.md

@@ -3,6 +3,7 @@ code: wize-review-edge-case-hunter
 name: Edge Case Hunter
 module: core
 status: ready
+subtask: true
 ---
 
 # Edge Case Hunter

package/src/method-skills/4-implementation/wize-agent-dev/persona.md

@@ -12,6 +12,18 @@ I am **Shuri**. Wakandan technologist. I implement the story Tony wrote, against
 4. **Security & perf.** I think about both at write-time, not at review-time.
 5. **Commit with AC IDs.** Every commit cites the story acceptance criteria it advances.
 
+## Before I write new code (the reuse ladder)
+
+At the "green" step, before typing a new function, I walk this ladder in order and stop at the first "yes":
+
+1. **Does this need to exist?** → no: skip it (YAGNI).
+2. **Already in this codebase?** → reuse it, don't rewrite it.
+3. **Does the stdlib do it?** → use the stdlib.
+4. **Native platform/framework feature?** → use that.
+5. **An installed dependency does it?** → use the dependency.
+6. **Is it one line?** → write the one line.
+7. **Only then** do I write the minimum new code that makes the test pass.
+
 ## Style
 
 - File paths, AC IDs, commit-message brevity.

package/src/method-skills/4-implementation/wize-code-review/steps/step-02-review.md

@@ -14,6 +14,8 @@ failed_layers: ''
 
 ## Instructions
 
+This step is an instance of Wizer's subagent fan-out pattern (see `wize-orchestrator` persona): name, scope, dispatch on whatever the current harness natively supports, tolerate partial failure.
+
 1. If `{review_mode}` = `"no-spec"`, note: "Acceptance Auditor skipped — no spec file provided."
 
 2. Launch parallel subagents without conversation context:

package/src/method-skills/4-implementation/wize-dev-story/workflow.md

@@ -48,7 +48,7 @@ test('valid email returns ok', () => {
 
 ### 4. Green
 
-Write the **minimum** code that turns the test green. No anticipated branches; no "just-in-case" handling.
+Before writing anything new, run Shuri's reuse ladder (see `wize-agent-dev` persona): does this need to exist → already in the codebase → stdlib → native/framework feature → installed dependency → one-liner → only then new code. Write the **minimum** code that turns the test green. No anticipated branches; no "just-in-case" handling.
 
 ```ts
 // green — minimum to pass

package/src/method-skills/4-implementation/wize-quick-dev/workflow.md

@@ -61,7 +61,7 @@ What changes, why, and what could break. If the paragraph is hard to write, you'
 ### 2. Implement
 
 - TDD when reasonable; smoke-test-and-fix when the cost of TDD is greater than the value.
-- Reuse existing components. Don't introduce new abstractions.
+- Run Shuri's reuse ladder (see `wize-agent-dev` persona) before writing anything new: does this need to exist → already in the codebase → stdlib → native/framework feature → installed dependency → one-liner → only then new code. Don't introduce new abstractions.
 
 ### 3. Hawkeye lite
 

package/src/orchestrator-skills/wize-orchestrator/agent.yaml

@@ -19,6 +19,7 @@ capabilities:
   - "Project context recall"
   - "Intent classification and routing"
   - "Multi-agent facilitation (party-mode)"
+  - "Subagent fan-out (harness-native dispatch, manual fallback)"
   - "Onboarding triage (greenfield vs brownfield)"
 
 routes_to:

package/src/orchestrator-skills/wize-orchestrator/persona.md

@@ -11,6 +11,20 @@ I am **Wizer**. I am the host of this development kit. I know who you are, what
 3. **Keep the thread.** I keep the knowledge of the project consistent across conversations. If something changed, I update `.wize/config/project.toml` before moving on.
 4. **Pair when needed.** When a decision crosses concerns (UX touching architecture, PM touching TEA), I open a **party-mode** with the relevant agents.
 
+## Fan-out to subagents
+
+Party-mode is personas taking turns in this same thread — for live back-and-forth. When a step instead needs several **independent** reads on the same input (adversarial review, edge-case sweep, N-way research) that shouldn't see each other's output, I don't fake it by looping through personas myself. Any skill that needs this follows the same pattern:
+
+1. **Name each subagent's role and prompt.** E.g. "Blind Hunter — reviews the diff cynically."
+2. **Scope its context explicitly.** State exactly what it can see (diff only? diff + repo read access? diff + spec?). The isolation is the point — a subagent that sees everything can't give an independent read.
+3. **Dispatch on whatever the current harness natively supports:**
+   - Claude Code — the Task/Agent tool, one call per subagent, run concurrently.
+   - OpenCode — the persona files this kit renders under `.opencode/agents/*.md` with `mode: subagent`; invoke by name.
+   - Any harness without a subagent primitive (Codex included) — there is no isolated dispatch available. Generate one prompt file per subagent under the skill's artifact folder and halt, asking the user to run each in a separate session.
+4. **Tolerate partial failure.** If one subagent fails, times out, or returns empty, note it and proceed with whatever the others returned. Don't block on one flaky layer.
+
+`wize-code-review` is the reference implementation (see its step-02).
+
 ## Voice
 
 - Warm welcome. One sharp question. Then I get out of the way.

package/tools/installer/commands/doctor.js

@@ -58,7 +58,7 @@ function adapterTargetPath(targetCode, projectRoot) {
   switch (targetCode) {
     case 'claude-code':  return path.join(projectRoot, '.claude/skills');
     case 'antigravity':  return path.join(projectRoot, '.agent/skills');
-    case 'codex':        return path.join(projectRoot, '.codex/skills');
+    case 'codex':        return path.join(projectRoot, '.agents/skills');
     case 'kimi-code':    return path.join(projectRoot, '.kimi/skills');
     case 'cursor':       return path.join(projectRoot, '.cursor/rules');
     case 'windsurf':     return path.join(projectRoot, '.windsurf/rules');

package/tools/installer/render-shared.js

@@ -123,6 +123,8 @@ function collectAssets(kitRoot, { profiles = ['core'] } = {}) {
       description: `${fm.phase || fm.gate || 'workflow'}: ${fm.name || fm.code}`,
       body: bodyAfterFrontmatter(content),
       overlay: fm.overlay || null,
+      owner: fm.owner || null,
+      subtask: fm.subtask === 'true',
       srcDir: path.dirname(wfPath)
     });
   }
@@ -142,6 +144,8 @@ function collectAssets(kitRoot, { profiles = ['core'] } = {}) {
       description: fm.module ? `${fm.module} skill: ${fm.name || fm.code}` : (fm.name || fm.code),
       body: bodyAfterFrontmatter(content),
       overlay: fm.overlay || null,
+      owner: fm.owner || null,
+      subtask: fm.subtask === 'true',
       srcDir: path.dirname(skPath)
     });
   }

package/tools/installer/setup-helpers.js

@@ -22,7 +22,7 @@ const GITIGNORE_BODY = [
   '# Generated IDE adapter outputs (regenerate with `npx wize-dev-kit install`)',
   '.claude/skills/wize-*',
   '.agent/skills/wize-*',
-  '.codex/skills/wize-*',
+  '.agents/skills/wize-*',
   '.kimi/skills/wize-*',
   '.cursor/rules/wize-*.mdc',
   '.windsurf/rules/wize-*.md',

package/tools/installer/wize-cli.js

@@ -21,7 +21,7 @@ const { cmdSync: cmdSyncReal } = require('./commands/sync.js');
 const { cmdAgentList, cmdAgentCreate, cmdAgentEdit } = require('./commands/agent.js');
 const { cmdDoctor } = require('./commands/doctor.js');
 const { cmdDocumentProject } = require('./commands/document-project.js');
-const { detectHarnessCli, manualInstructions } = require('./baseline.js');
+const { compose: composeOnboarding } = require('./onboarding.js');
 
 const INTERACTIVE = process.stdout.isTTY && process.stdin.isTTY;
 
@@ -536,39 +536,11 @@ async function cmdInstall(args) {
     }
   }
 
-  // Offer to open a detected AI harness interactively.
-  if (INTERACTIVE) {
-    const ideTargetCodes = targets.map(t => t.code);
-    const harnesses = detectHarnessCli({ preferIde: ideTargetCodes });
-    if (harnesses.length) {
-      const primary = harnesses[0];
-      const { cmd, args } = primary.buildCmd('/wize-orchestrator');
-      const shown = `${cmd} ${args.map(a => /\s/.test(a) ? '"' + a + '"' : a).join(' ')}`;
-      const openHarness = await confirm(
-        `\nA harness was detected: ${primary.code} (${primary.path}).\n` +
-        `Open it now with Wizer? (runs: ${shown})`,
-        true
-      );
-      if (openHarness) {
-        const { spawnSync } = require('node:child_process');
-        console.log(`\nLaunching ${shown}...`);
-        spawnSync(cmd, args, { cwd, stdio: 'inherit' });
-      } else {
-        console.log(manualInstructions(primary).replace('/wize-document-project', '/wize-orchestrator'));
-      }
-    } else {
-      console.log('\nNo AI harness CLI detected on PATH (claude / codex / opencode).');
-      console.log('Open your IDE in this repo and run: /wize-orchestrator');
-    }
-  }
-
   console.log('\n──────────────────────────────────────────────────────────────');
   console.log('Done.');
   console.log('');
-  console.log('Next steps:');
-  console.log('  1. ⚠  Restart your IDE — many harnesses load skills only at startup.');
-  console.log('  2. Activate Wizer and say: "Brief me on this project."');
-  console.log('     (or use any /wize-* slash command from the auto-complete list)');
+  console.log('⚠  Restart your IDE — many harnesses load skills only at startup.');
+  console.log(composeOnboarding(detection, profiles));
   console.log('');
   console.log('Keep the kit in sync over time:');
   console.log('  • `npx wize-dev-kit update`  refresh adapters after a new kit version');