crewkit 1.1.5 → 1.1.6

package/README.md

@@ -94,10 +94,9 @@ npx crewkit add impact         # analise de blast radius
 ## IDEs suportadas
 
 - **Claude Code** — suporte completo (agents, skills, hooks, rules, memory)
-- **GitHub Copilot** — auto-detectado, gera `copilot-instructions.md`, agents e prompts
-- **Cursor** — auto-detectado, gera rules com globs e `AGENTS.md`
+- **GitHub Copilot** — gera `copilot-instructions.md`, agents, skills nativas, prompts, instructions por stack e guardrails de seguranca
 
-A IDE e detectada automaticamente durante o setup. `.ai/memory/` e compartilhado entre todas as IDEs.
+A IDE e selecionada pelo usuario durante o setup. `.ai/memory/` e compartilhado entre as IDEs.
 
 ## O que e commitado vs ignorado
 
@@ -111,7 +110,7 @@ A IDE e detectada automaticamente durante o setup. `.ai/memory/` e compartilhado
 ## Requisitos
 
 - Node.js >= 20
-- Claude Code, GitHub Copilot ou Cursor
+- Claude Code ou GitHub Copilot
 - Opus ou Sonnet recomendado (Haiku funciona mas gera scans mais rasos)
 
 ## Licenca
@@ -216,10 +215,9 @@ npx crewkit add impact         # blast radius analysis
 ## Supported IDEs
 
 - **Claude Code** — full support (agents, skills, hooks, rules, memory)
-- **GitHub Copilot** — auto-detected, generates `copilot-instructions.md`, agents and prompts
-- **Cursor** — auto-detected, generates rules with globs and `AGENTS.md`
+- **GitHub Copilot** — generates `copilot-instructions.md`, agents, native skills, prompts, per-stack instructions and security guardrails
 
-IDE is auto-detected during setup. `.ai/memory/` is shared across all IDEs.
+IDE is selected by the user during setup. `.ai/memory/` is shared across IDEs.
 
 ## What gets committed vs ignored
 
@@ -233,7 +231,7 @@ IDE is auto-detected during setup. `.ai/memory/` is shared across all IDEs.
 ## Requirements
 
 - Node.js >= 20
-- Claude Code, GitHub Copilot, or Cursor
+- Claude Code or GitHub Copilot
 - Opus or Sonnet recommended (Haiku works but produces shallower scans)
 
 ## License

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "crewkit",
-  "version": "1.1.5",
+  "version": "1.1.6",
   "description": "Context engineering framework for AI-assisted development. One command to set up agents, skills, hooks, rules, and memory for your project.",
   "type": "module",
   "bin": {
@@ -20,7 +20,6 @@
     "ai",
     "claude-code",
     "copilot",
-    "cursor",
     "agents",
     "skills",
     "mcp",

package/skill/SKILL.md

@@ -15,7 +15,7 @@ This skill scans the current project's codebase and generates a complete, calibr
 
 ### Critical rules (read before starting)
 1. **Language:** All generated files MUST be in **English**. User-facing output (reports during execution) matches the user's language.
-2. **Zero questions:** Do not pause for user input except for re-run detection ([R]/[M]/[C]).
+2. **Zero questions:** Do not pause for user input except for re-run detection ([R]/[M]/[C]) and IDE target selection.
 3. **Validate everything:** After all files are generated, run the completion checklist at the bottom of this document before reporting done.
 4. **Persist to disk:** Write scan results to `.crewkit/` files at each phase. Read them back before generating. This survives context compaction.
 
@@ -40,6 +40,21 @@ Check if `.crewkit/version` exists.
 
 Wait for user choice. On [R] or [M]: existing `lessons-*.md` files with content are **preserved and merged**, not overwritten — these contain accumulated institutional knowledge.
 
+### 3. IDE target selection
+
+**Always ask the user which IDEs to generate context for.**
+
+Check if a previous selection exists in `.crewkit/scan-phase1-recon.md`. If it does, show it as the default.
+
+> "Which IDEs should I generate context for?"
+> - **[1] Claude Code**
+> - **[2] GitHub Copilot**
+> - **[A] All of the above**
+>
+> You can select multiple (e.g., "1,2"). Default: [previous selection if re-run, or "1" if first run].
+
+Accept the user's choice. If the user presses Enter without typing, use the default.
+
 ---
 
 ## Phase 0 — Preparation
@@ -51,7 +66,7 @@ If `.claude/` or `CLAUDE.md` or `.ai/memory/` exists:
 3. Report: "Existing setup backed up to `.crewkit-backup/`"
 
 ### Create directory structure
-Create directories based on the **selected IDE targets** from Phase 1:
+Create directories based on the **selected IDE targets** from Pre-flight:
 
 **Always:**
 ```
@@ -75,12 +90,8 @@ Create directories based on the **selected IDE targets** from Phase 1:
 .github/
   agents/
   instructions/
-```
-
-**If Cursor is a selected target:**
-```
-.cursor/
-  rules/
+  skills/
+  prompts/
 ```
 
 ### Version tracking
@@ -162,35 +173,11 @@ For each stack, detect the build and test commands:
 
 Read `package.json` scripts, CI configs, or Makefiles to find the actual commands used in this project. Prefer project-specific over generic.
 
-### IDE Target Detection
-
-Detect which IDEs are available. Check **two sources**: project-level signals AND global install markers.
+### IDE targets
 
-| IDE | Project signals | Global install marker |
-|-----|----------------|----------------------|
-| Claude Code | `.claude/` directory exists | `~/.claude/skills/crewkit-setup/` exists |
-| GitHub Copilot | `.github/` directory exists OR `.github/copilot-instructions.md` found | `~/.copilot/agents/crewkit-setup.md` exists |
-| Cursor | `.cursor/` directory exists OR `.cursorrules` file found | `~/.cursor/crewkit-setup.md` exists |
-
-An IDE is **available** if ANY of its signals match (project OR global).
-
-**IDE target selection (exception to zero-questions rule):**
-
-If more than one IDE is available, present the user with a choice:
-
-> "Multiple IDEs detected. Generate context for which?"
-> - **[1] Claude Code**
-> - **[2] GitHub Copilot**
-> - **[3] Cursor**
-> - **[4] All detected**
->
-> (Only show IDEs that were detected. Default: all detected.)
-
-If only one IDE is available, use it without asking.
-
-Save the **selected** IDE targets to scan data. Example:
+Use the IDE targets selected during Pre-flight (step 3). Include them in the scan data:
 ```
-IDE targets selected: GitHub Copilot
+IDE targets selected: [selected IDEs from pre-flight]
 ```
 
 **Store results as: `ReconProfile`** — write to `.crewkit/scan-phase1-recon.md` for persistence.
@@ -214,10 +201,8 @@ Read existing human-written documentation to understand project context, convent
 **Existing AI context (from any IDE — preserve, don't discard):**
 - Existing `CLAUDE.md` (if backed up from a previous setup)
 - Existing `.ai/memory/` files (if backed up)
-- `.cursorrules` or `.cursor/rules/*.md` (Cursor)
 - `.github/copilot-instructions.md` or `.github/instructions/*.md` (GitHub Copilot)
 - `.github/agents/*.agent.md` (Copilot agents)
-- `AGENTS.md` (Cursor agents)
 
 If any of these exist, read them and extract rules, conventions, and patterns to incorporate into the generated setup. This preserves context engineering work done in other IDEs.
 
@@ -446,7 +431,7 @@ Compile all profiles into a single summary and present it to the user.
 - Dev server: `[command]` (if detected)
 
 ## IDE Targets (selected)
-- [list selected IDE targets — e.g., "GitHub Copilot" or "Claude Code, Cursor"]
+- [list selected IDE targets — e.g., "Claude Code" or "Claude Code, GitHub Copilot"]
 ```
 
 ### Save profile
@@ -480,9 +465,9 @@ Generate files based on the **selected IDE targets** from Phase 1. The generatio
 - **Steps 2-9** (CLAUDE.md, `.claude/agents/`, `.claude/rules/`, `.claude/settings.json`, `.claude/hooks/`, `.claude/skills/`, `.claude/QUICKSTART.md`, `.mcp.json`): ONLY if **Claude Code** is a selected target.
 - **Step 10** (IDE Adapters): Runs for each non-Claude IDE in the selected targets. Adapters generate directly from scan data + `.ai/memory/` — they do NOT depend on Claude files existing.
 
-If the only selected target is GitHub Copilot or Cursor, Steps 2-9 are **skipped entirely**. The adapter generates the IDE-native files directly.
+If the only selected target is GitHub Copilot, Steps 2-9 are **skipped entirely**. The adapter generates the IDE-native files directly.
 
-**If the user chose [M] (memory only):** generate Step 1 (memory) + the project rules file for the selected IDE (CLAUDE.md / copilot-instructions.md / .cursor/rules/project.md), then update context headers in existing agents. Stop after that.
+**If the user chose [M] (memory only):** generate Step 1 (memory) + the project rules file for the selected IDE (CLAUDE.md / copilot-instructions.md), then update context headers in existing agents. Stop after that.
 
 ### Templates directory
 Templates are located at `~/.claude/skills/crewkit-setup/templates/`. Read them from disk.
@@ -953,7 +938,6 @@ Read all core skill templates from `~/.claude/skills/crewkit-setup/templates/ski
 - `hotfix/SKILL.md`
 - `explore-and-plan/SKILL.md`
 - `review-pr/SKILL.md`
-
 Copy each skill template to `.claude/skills/[name]/SKILL.md`. **If the template has a `references/` subdirectory, copy it too.**
 
 These skill templates are **stack-agnostic** by design — they reference `.ai/memory/commands.md` for build/test commands and `.ai/memory/` for project context. No variable substitution needed.
@@ -980,6 +964,7 @@ This project uses crewkit for AI-assisted development. Here's how to use it.
 - `/explore-and-plan <feature>` — map a module and plan before coding (recommended for LARGE tasks)
 - `/review-pr [number]` — review a pull request
 
+
 ## How agents work together
 The orchestrator routes work to specialized agents:
 - **explorer** (Sonnet) — finds files, maps dependencies — read-only
@@ -1046,14 +1031,11 @@ For any placeholder tokens, add a comment in the profile or tell the user to fil
 
 ### Step 10 — IDE Adapters
 
-For each **non-Claude IDE** in the selected targets (GitHub Copilot, Cursor):
+If **GitHub Copilot** is a selected target:
 
-1. For each selected non-Claude IDE target:
-   a. Determine the adapter file path: `~/.claude/skills/crewkit-setup/adapters/{ide}.md`
-      - GitHub Copilot → `adapters/copilot.md`
-      - Cursor → `adapters/cursor.md`
-   b. If the adapter file does not exist: WARN the user ("Adapter file not found for [IDE] — skipping") and continue to the next IDE. Do not fail.
-   c. If the adapter file exists: read it completely, then MANDATORY: follow ALL instructions in it.
+1. Determine the adapter file path: `~/.claude/skills/crewkit-setup/adapters/copilot.md`
+2. If the adapter file does not exist: WARN the user ("Adapter file not found for Copilot — skipping"). Do not fail.
+3. If the adapter file exists: read it completely, then MANDATORY: follow ALL instructions in it.
 
 **Source of truth for adapters (in priority order):**
 1. If Claude Code files exist (Steps 2-9 were generated): use them as the primary source, converting content to the target IDE format.
@@ -1089,14 +1071,11 @@ After all generation steps, run the **Completion Checklist** (at the bottom of t
 - `.mcp.json` — [N] MCP servers
 
 ## Generated files (GitHub Copilot) — only if selected
-- `.github/copilot-instructions.md` — project rules
-- `.github/instructions/` — [N] per-stack instruction files
-- `.github/agents/` — 5 agent files
-
-## Generated files (Cursor) — only if selected
-- `.cursor/rules/project.md` — project rules
-- `.cursor/rules/` — [N] per-stack rule files
-- `AGENTS.md` — all agents
+- `.github/copilot-instructions.md` — project rules (no frontmatter, includes session start + quality gate + safety rules)
+- `.github/instructions/` — [N] per-stack instruction files + sensitive-files guardrail
+- `.github/agents/` — 5 agent files (with model, tools, mcp-servers)
+- `.github/skills/` — 4 native skills
+- `.github/prompts/` — 4 IDE fallback prompts
 
 ## Validation: [X]/[Y] checks passed
 
@@ -1108,7 +1087,6 @@ After all generation steps, run the **Completion Checklist** (at the bottom of t
 [Adapt next steps to the selected IDE:]
 - Claude Code: "Review CLAUDE.md → run /full-workflow"
 - Copilot: "Review .github/copilot-instructions.md → open Copilot Chat"
-- Cursor: "Review .cursor/rules/project.md → start coding"
 - All: "Commit the setup"
 ```
 
@@ -1144,17 +1122,16 @@ Before presenting the Final Report, go through EVERY item. Fix failures before r
 - [ ] `.mcp.json` — read back, verify valid JSON, has context7
 - [ ] `.claude/hooks/*.sh` — run `bash -n` on each, all 4 pass syntax check
 - [ ] `.claude/rules/` — at least 1 per detected stack, glob patterns match real files in project
-- [ ] `.claude/skills/` — all 4 core skills have SKILL.md
+- [ ] `.claude/skills/` — all 4 core skills have SKILL.md (full-workflow, hotfix, explore-and-plan, review-pr)
 
 ### Content checks (Copilot — only if Copilot is a selected target)
-- [ ] `.github/copilot-instructions.md` — exists, contains project hard rules, no Claude-specific sections
+- [ ] `.github/copilot-instructions.md` — exists, **NO frontmatter**, contains hard rules + session start + quality gate + safety rules, no Claude-specific sections
 - [ ] `.github/instructions/` — at least 1 per detected stack with `applyTo:` frontmatter
-- [ ] `.github/agents/` — 5 agent files with `.agent.md` extension, no `model:` lines
-
-### Content checks (Cursor — only if Cursor is a selected target)
-- [ ] `.cursor/rules/project.md` — exists, has `alwaysApply: true` frontmatter, no Claude-specific sections
-- [ ] `.cursor/rules/` — at least 1 per detected stack with `globs:` frontmatter
-- [ ] `AGENTS.md` — exists at project root with all 5 agents as `##` sections
+- [ ] `.github/instructions/sensitive-files.instructions.md` — exists with glob for `.env*`, `credentials*`, `secrets*`, `*.key`, `*.pem`
+- [ ] `.github/agents/` — 5 agent files with `.agent.md` extension, each has `tools:` with canonical aliases (`read`, `edit`, `search`, `execute`), has `model:` converted to Copilot names, body under 30,000 chars
+- [ ] `.github/agents/coder.agent.md` + `tester.agent.md` — have `mcp-servers:` if `.mcp.json` exists
+- [ ] `.github/skills/` — 4 native skill files with `name:` + `description:` frontmatter
+- [ ] `.github/prompts/` — 4 IDE fallback prompts with `description:` + `agent: "agent"` frontmatter
 
 ### Integrity checks (always)
 - [ ] `.crewkit/last-scan.md` — exists with full profile including Domain section

package/skill/adapters/copilot.md

@@ -15,8 +15,8 @@ You are the AI. Follow every instruction in this file to generate GitHub Copilot
 
 1. All generated files MUST be in **English**.
 2. Do NOT duplicate `.ai/memory/` — it is shared between all IDEs. No transformation needed.
-3. `model:` frontmatter from `.claude/agents/*.md` is Claude Code-only — strip it.
-4. Skills are lossy when converted to Copilot prompts — extract workflow intent only, drop multi-agent orchestration mechanics.
+3. Agent `model:` from `.claude/agents/*.md` MUST be preserved and converted to Copilot model names (see Step C3).
+4. Skills can be converted **natively** to `.github/skills/` (not lossy). Prompt files (`.prompt.md`) are generated as an additional IDE-only fallback.
 5. Create `.github/` subdirectories if they do not exist.
 
 ---
@@ -24,12 +24,37 @@ You are the AI. Follow every instruction in this file to generate GitHub Copilot
 ## Step C1 — `.github/copilot-instructions.md`
 
 **Source:** `CLAUDE.md` if it exists, otherwise `.crewkit/last-scan.md` + `.ai/memory/conventions.md`
-**Transformation:** If using CLAUDE.md, reformat for Copilot — remove Agent Discipline, Skills (slash commands), Architect Decision Gate, and Test Safety Loop sections (Claude Code-specific). Keep hard rules, overview, memory loading instructions, and output format. If generating from scan data, create the content directly using the project profile and conventions.
+
+**IMPORTANT:** This file has **NO frontmatter**. It is plain Markdown only.
+
+**Transformation:** If using CLAUDE.md, reformat for Copilot — remove Agent Discipline, Skills (slash commands), Architect Decision Gate, and Test Safety Loop sections (Claude Code-specific). If generating from scan data, create the content directly using the project profile and conventions.
+
+**What to INCLUDE (absorb from files that have no Copilot equivalent):**
+
+From `CLAUDE.md`:
+- Overview, hard rules, memory loading instructions, output format
+
+From `.claude/QUICKSTART.md` (onboarding — no Copilot equivalent):
+- Day-to-day workflow section (which agents to use for what)
+
+From `.claude/hooks/session-start.sh` (no Copilot equivalent):
+- "At the start of every conversation, read `.ai/memory/architecture.md` and `.ai/memory/conventions.md`. Check `git log --oneline -5` for recent work."
+
+From `.claude/hooks/stop-quality-gate.sh` (no Copilot equivalent):
+- "Before completing any task, run build and test commands from `.ai/memory/commands.md`. Never report success with failing tests or broken builds."
+
+From `.claude/napkin.md` (no Copilot equivalent):
+- "Read `.claude/napkin.md` at the start of every session for current priorities and blockers."
+
+From `.claude/settings.json` deny list (no Copilot equivalent):
+- "Never execute destructive commands: `rm -rf`, `sudo`, `DROP DATABASE`, `git push --force`."
 
 **Expected output format:**
 ```markdown
 # [PROJECT NAME] — Copilot Instructions
 
+> These instructions apply to all GitHub Copilot interactions in this repository.
+
 ## Overview
 [1-2 sentences — what the project is, main stack]
 [Business domain: what it does, core entities, risk profile]
@@ -51,6 +76,32 @@ Details for each rule → `.ai/memory/conventions.md`
 
 ---
 
+## Session start
+
+At the start of every conversation:
+1. Read `.ai/memory/architecture.md` and `.ai/memory/conventions.md`
+2. Read `.claude/napkin.md` for current priorities and blockers
+3. Check `git log --oneline -5` and `git status` for recent work context
+
+---
+
+## Quality gate
+
+Before completing any task:
+1. Run the build command: `[detected build command]`
+2. Run the test command: `[detected test command]`
+3. Never report success with failing tests or broken builds
+
+---
+
+## Safety rules
+
+- Never edit sensitive files: `.env`, `.env.*`, `credentials.*`, `secrets.*`, `*.key`, `*.pem`
+- Never execute destructive commands: `rm -rf`, `sudo`, `DROP DATABASE`, `git push --force`
+- Move secrets to environment variables, never hardcode them
+
+---
+
 ## Project Memory (`.ai/memory/`)
 
 Load context on demand — do not load all files every time:
@@ -65,6 +116,16 @@ Load context on demand — do not load all files every time:
 
 ---
 
+## Workflow
+
+- Use `@explorer` to find files, map dependencies — read-only
+- Use `@architect` to evaluate design options — read-only
+- Use `@coder` to implement the smallest correct diff
+- Use `@tester` to create tests, run full suite
+- Use `@reviewer` to review code — find real bugs, no noise
+
+---
+
 ## Output Format
 
 Always return:
@@ -72,23 +133,22 @@ Always return:
 - **Files changed** — list with brief description
 - **Tests** — pass/fail count (if tests were run)
 - **Risks / Next steps** — if any
+
+> Always think step by step. Never report success with failing tests.
 ```
 
 **What to REMOVE from CLAUDE.md:**
 - `## Agent Discipline` section (orchestrator/worker model is Claude Code-only)
 - `## Skills (slash commands)` section (slash commands are Claude Code-only)
 - `## Architect Decision Gate` section
-- `## Test Safety Loop` section (keep the intent as a single sentence in the output format instead)
-
-**What to ADD:**
-- At the top, after the title: `> These instructions apply to all GitHub Copilot interactions in this repository.`
-- In the output format section: `> Always think step by step. Never report success with failing tests.`
+- `## Test Safety Loop` section (absorbed into Quality gate section above)
 
 ---
 
 ## Step C2 — `.github/instructions/*.instructions.md`
 
 **Source:** `.claude/rules/*.md` if they exist, otherwise generate directly from scan data (detected stacks + HIGH/MEDIUM confidence patterns).
+
 **Transformation:** If using Claude rules, convert frontmatter to Copilot format — keep glob patterns and rule content unchanged. If generating from scan data, create one instructions file per detected stack with appropriate glob patterns and rules.
 
 Claude Code frontmatter format:
@@ -99,7 +159,7 @@ globs: "src/**/*.{js,ts}"
 ---
 ```
 
-Copilot instructions format:
+Copilot instructions format (**no `description:` in frontmatter**):
 ```markdown
 ---
 applyTo: "src/**/*.{js,ts}"
@@ -114,31 +174,29 @@ applyTo: "src/**/*.{js,ts}"
 - `description:` → remove from frontmatter, use as the first `# heading` in the body instead
 - All rule content (body): copy verbatim
 
+**Optional frontmatter field:**
+- `excludeAgent:` — set to `"code-review"` or `"coding-agent"` to hide instructions from a specific Copilot agent. Omit to apply to all agents. Only add if a rule is clearly irrelevant to one agent type.
+
 **File naming:** `.claude/rules/dotnet.md` → `.github/instructions/dotnet.instructions.md`
 Strip any existing `.md` suffix and append `.instructions.md`.
 
-**Example — source `.claude/rules/node.md`:**
-```markdown
----
-description: "Node.js coding rules — applied when editing src/**/*.{js,ts}"
-globs: "src/**/*.{js,ts}"
----
-
-# Node.js Rules
+### Additional instructions file: sensitive files guardrail
 
-- Use async/await, not .then() chains
-- Validate all external input at the handler boundary
-```
+Generate an extra instructions file that replaces the `protect-sensitive-files.sh` hook:
 
-**Example — target `.github/instructions/node.instructions.md`:**
+**File:** `.github/instructions/sensitive-files.instructions.md`
 ```markdown
 ---
-applyTo: "src/**/*.{js,ts}"
+applyTo: "**/.env*,**/credentials*,**/secrets*,**/*.key,**/*.pem,**/appsettings.*.json"
 ---
-# Node.js Rules
+# Sensitive Files — Do Not Edit
 
-- Use async/await, not .then() chains
-- Validate all external input at the handler boundary
+These files contain secrets or environment-specific configuration.
+
+- NEVER modify these files directly
+- NEVER write secrets, API keys, or passwords into any file
+- Move all secrets to environment variables
+- If a secret is needed, reference it via `process.env.VAR_NAME` or the equivalent for the stack
 ```
 
 ---
@@ -146,43 +204,147 @@ applyTo: "src/**/*.{js,ts}"
 ## Step C3 — `.github/agents/*.agent.md`
 
 **Source:** `.claude/agents/*.md` if they exist, otherwise generate directly from the agent templates at `~/.claude/skills/crewkit-setup/templates/agents/` (or from scan data if templates are unavailable).
-**Transformation:** Strip `model:` frontmatter line. Keep `name:` and `description:`. Remove the `<!-- crewkit:context-start -->...<!-- crewkit:context-end -->` block — Copilot agents do not use this inline context injection. Keep the full agent role description and instructions. When generating from templates, inject project context from `.crewkit/last-scan.md` into each agent.
+
+**Transformation:** Convert `model:` to Copilot model names. Remove the `<!-- crewkit:context-start -->...<!-- crewkit:context-end -->` block. Add `tools:` and optionally `mcp-servers:` to frontmatter. Keep the full agent role description and instructions. When generating from templates, inject project context from `.crewkit/last-scan.md` into each agent.
 
 **Copilot agent frontmatter format:**
-```markdown
+```yaml
 ---
 name: explorer
 description: "Read-only reconnaissance agent. Maps files, dependencies, and caller chains."
 tools:
-  - read_file
-  - list_directory
-  - search_files
+  - read
+  - search
 ---
 ```
 
-**`tools:` mapping — add if applicable:**
-| Agent | Suggested Copilot tools |
-|-------|------------------------|
-| explorer | `read_file`, `list_directory`, `search_files` |
-| architect | `read_file`, `search_files` |
-| coder | `read_file`, `create_file`, `replace_string_in_file` |
-| tester | `read_file`, `create_file`, `run_in_terminal` |
-| reviewer | `read_file`, `search_files` |
+**Model conversion (preserve, don't remove):**
+
+| Claude Code `model:` | Copilot `model:` |
+|----------------------|-----------------|
+| `opus` | `"Claude Opus 4"` |
+| `sonnet` | `"Claude Sonnet 4"` |
+| (omitted) | (omit — uses default) |
+
+If the project uses multiple AI providers, use an array for fallback: `model: ["Claude Sonnet 4", "GPT-4o"]`
+
+**`tools:` mapping — use canonical Copilot tool aliases:**
+
+| Agent | tools | Rationale |
+|-------|-------|-----------|
+| explorer | `["read", "search"]` | Read-only reconnaissance |
+| architect | `["read", "search"]` | Read-only design review |
+| coder | `["read", "edit", "search", "execute"]` | Full implementation access |
+| tester | `["read", "edit", "search", "execute"]` | Test creation + execution |
+| reviewer | `["read", "search"]` | Read-only code review |
+
+**Available canonical tool aliases:** `read`, `edit`, `search`, `execute` (shell), `web` (search/fetch), `agent` (subagents), `todo` (task management). Use `["*"]` for all tools.
+
+**Optional: `mcp-servers:` field.** If the project uses MCP servers (detected from `.mcp.json`), add them to relevant agents (coder, tester):
+```yaml
+mcp-servers:
+  context7:
+    type: "local"
+    command: "npx"
+    args: ["-y", "@context7/mcp-server"]
+    tools: ["*"]
+  postgres:
+    type: "local"
+    command: "npx"
+    args: ["-y", "@modelcontextprotocol/server-postgres"]
+    tools: ["*"]
+    env:
+      DATABASE_URL: ${{ secrets.COPILOT_MCP_DATABASE_URL }}
+```
+
+Only add MCP servers to agents that need them (coder + tester for DB servers, all agents for context7).
+
+**Optional: `agents:` field.** To allow an agent to invoke other agents as subagents:
+```yaml
+agents: ["*"]   # allow invoking any other agent
+```
+Add this to `coder` (can call `explorer` for context) and `tester` (can call `coder` for fixes).
+
+**Body limit:** Maximum **30,000 characters** for Markdown content below frontmatter. If an agent body exceeds this, trim low-priority sections.
 
 **What to strip:**
-- `model:` line
 - The entire `<!-- crewkit:context-start -->...<!-- crewkit:context-end -->` block (inclusive)
 
+**What to ADD to the body (absorb from deny list — no Copilot equivalent):**
+- For `coder` and `tester` agents, add at the end of the body:
+  ```
+  ## Safety
+  - Never execute: `rm -rf`, `sudo`, `DROP DATABASE`, `git push --force`
+  - Never edit: `.env`, `credentials.*`, `secrets.*`, `*.key`, `*.pem`
+  ```
+
 **File naming:** `.claude/agents/explorer.md` → `.github/agents/explorer.agent.md`
 
 ---
 
-## Step C4 — `.github/prompts/*.prompt.md`
+## Step C4 — `.github/skills/*/SKILL.md` (native skills)
 
 **Source:** `.claude/skills/*/SKILL.md`
-**Transformation:** LOSSY. Extract the workflow intent as single-shot guidance. Drop multi-agent orchestration, agent routing, fix loops, and exit gates — Copilot prompts are single-turn, not multi-agent pipelines.
 
-**Extract:**
+**Transformation:** Copilot supports skills natively with a compatible format. Copy the skill content and adapt the frontmatter.
+
+**Copilot skill frontmatter format:**
+```yaml
+---
+name: "full-workflow"
+description: "Execute the complete development workflow: explore, implement, test, review."
+user-invocable: true
+---
+```
+
+**Frontmatter fields:**
+- `name:` (required) — lowercase with hyphens, max 64 characters
+- `description:` (required) — what the skill does and when to use it, max 1024 characters
+- `user-invocable:` (optional) — default `true`, set to `false` for internal-only skills
+- `allowed-tools:` (optional) — list of tools to pre-approve (e.g., `["execute"]`)
+
+**Body adaptation:**
+- Keep the workflow steps, classification tables, and output format
+- Replace Claude Code-specific references:
+  - `/compact` → remove (Copilot has no context compaction)
+  - `→ explorer` agent routing → `@explorer` mention syntax
+  - References to `.claude/settings.json` → remove
+- Keep references to `.ai/memory/` (Copilot reads these)
+- Keep build/test command references to `.ai/memory/commands.md`
+
+**File naming:** `.claude/skills/full-workflow/SKILL.md` → `.github/skills/full-workflow/SKILL.md`
+If the skill has a `references/` subdirectory, copy it too.
+
+---
+
+## Step C5 — `.github/prompts/*.prompt.md` (IDE fallback)
+
+**Source:** `.claude/skills/*/SKILL.md`
+
+**Transformation:** LOSSY. These are a simplified fallback for IDEs that don't support `.github/skills/`. Extract the workflow intent as single-shot guidance.
+
+**Copilot prompt frontmatter format:**
+```yaml
+---
+description: "Complete development workflow: explore, implement, test, and review."
+agent: "agent"
+tools:
+  - read
+  - edit
+  - search
+  - execute
+---
+```
+
+**Frontmatter fields:**
+- `description:` (optional) — what the prompt does, shown in UI
+- `agent:` (optional) — `"agent"` (full tool use), `"ask"` (no tools), `"plan"`, or a custom agent name
+- `tools:` (optional) — available tools during execution
+- `model:` (optional) — specific model to use
+- `argument-hint:` (optional) — hint text shown in chat input
+- Do NOT use `name:` — the prompt name is derived from the filename
+
+**Extract from skill:**
 - What the skill is for (description)
 - The key steps the user should follow or the AI should perform
 - Any output format requirements
@@ -194,16 +356,16 @@ tools:
 - Phase breakdown with parallel agents
 - Any reference to Claude Code slash commands (`/compact`, `/crewkit-setup`)
 
-**Example — source `full-workflow/SKILL.md` intent:**
-```
-Routes tasks through explore → implement → test → review agents.
-```
-
 **Example — target `.github/prompts/full-workflow.prompt.md`:**
 ```markdown
 ---
-name: full-workflow
 description: "Complete development workflow: explore, implement, test, and review."
+agent: "agent"
+tools:
+  - read
+  - edit
+  - search
+  - execute
 ---
 # Full Workflow
 
@@ -223,7 +385,7 @@ Use this prompt to complete a development task end-to-end.
 ```
 
 **File naming:** `full-workflow/SKILL.md` → `.github/prompts/full-workflow.prompt.md`
-Use the skill directory name as the prompt file name.
+Use the skill directory name as the prompt file name. The user invokes it in Copilot Chat as `/full-workflow`.
 
 ---
 
@@ -231,12 +393,13 @@ Use the skill directory name as the prompt file name.
 
 Before reporting done, verify each item:
 
-- [ ] `.github/copilot-instructions.md` — exists, contains hard rules, does NOT contain Agent Discipline or slash command sections
-- [ ] `.github/instructions/` — one `.instructions.md` file per `.claude/rules/*.md` source file
-- [ ] `.github/instructions/*.instructions.md` — each has `applyTo:` frontmatter (not `globs:`)
-- [ ] `.github/agents/` — one `.agent.md` per `.claude/agents/*.md` source file
-- [ ] `.github/agents/*.agent.md` — no `model:` line, no `crewkit:context-start` block, has `tools:` frontmatter
-- [ ] `.github/prompts/` — one `.prompt.md` per `.claude/skills/*/SKILL.md` source file
-- [ ] `.github/prompts/*.prompt.md` — each has `name:` and `description:` frontmatter, no multi-agent routing language
+- [ ] `.github/copilot-instructions.md` — exists, **NO frontmatter**, contains hard rules, session start, quality gate, safety rules. Does NOT contain Agent Discipline or slash command sections
+- [ ] `.github/instructions/` — one `.instructions.md` file per detected stack with `applyTo:` frontmatter
+- [ ] `.github/instructions/sensitive-files.instructions.md` — exists with glob for `.env*`, `credentials*`, `secrets*`, `*.key`, `*.pem`
+- [ ] `.github/agents/` — 5 agent files with `.agent.md` extension
+- [ ] `.github/agents/*.agent.md` — each has `tools:` with canonical aliases (`read`, `edit`, `search`, `execute`), has `model:` converted to Copilot names, no `crewkit:context-start` block, body under 30,000 chars
+- [ ] `.github/agents/coder.agent.md` + `tester.agent.md` — have `mcp-servers:` if `.mcp.json` exists
+- [ ] `.github/skills/` — one `SKILL.md` per `.claude/skills/*/SKILL.md` source, with `name:` + `description:` frontmatter
+- [ ] `.github/prompts/` — one `.prompt.md` per skill as IDE fallback, with `description:` + `agent: "agent"` frontmatter
 - [ ] `.ai/memory/` — NOT duplicated under `.github/` (shared, no copy needed)
 - [ ] No Portuguese in any generated file

package/skill/copilot-agent.md

@@ -1,6 +1,6 @@
 ---
 name: crewkit-setup
-description: "Scan your codebase and generate a complete context engineering setup: agents, instructions, memory files, and copilot-instructions.md."
+description: "Scan your codebase and generate a complete context engineering setup: agents, skills, instructions, prompts, memory files, and copilot-instructions.md."
 ---
 
 # crewkit-setup
@@ -63,9 +63,13 @@ Create these files (all in English):
 
 ### Step 2 — `.github/copilot-instructions.md`
 
+**IMPORTANT: No frontmatter. Plain Markdown only.**
+
 ```markdown
 # [PROJECT NAME] — Copilot Instructions
 
+> These instructions apply to all GitHub Copilot interactions in this repository.
+
 ## Overview
 [1-2 sentences: what the project is, main stack]
 
@@ -82,12 +86,49 @@ Details → `.ai/memory/conventions.md`
 
 ---
 
+## Session start
+
+At the start of every conversation:
+1. Read `.ai/memory/architecture.md` and `.ai/memory/conventions.md`
+2. Check `git log --oneline -5` for recent work context
+
+---
+
+## Quality gate
+
+Before completing any task:
+1. Run the build command: `[detected build command]`
+2. Run the test command: `[detected test command]`
+3. Never report success with failing tests or broken builds
+
+---
+
+## Safety rules
+
+- Never edit sensitive files: `.env`, `.env.*`, `credentials.*`, `secrets.*`, `*.key`, `*.pem`
+- Never execute destructive commands: `rm -rf`, `sudo`, `DROP DATABASE`, `git push --force`
+
+---
+
 ## Memory
 
 - `.ai/memory/architecture.md` — system design
 - `.ai/memory/conventions.md` — coding conventions
 - `.ai/memory/commands.md` — build/test commands
+- `.ai/memory/testing.md` — test strategy
 - `.ai/memory/lessons.md` — known gotchas
+
+---
+
+## Output Format
+
+Always return:
+- **Summary** — what was done
+- **Files changed** — list with brief description
+- **Tests** — pass/fail count (if tests were run)
+- **Risks / Next steps** — if any
+
+> Always think step by step. Never report success with failing tests.
 ```
 
 ### Step 3 — `.github/instructions/` (per-stack rules)
@@ -97,25 +138,85 @@ For each detected stack, create `[stack].instructions.md`:
 ---
 applyTo: "**/*.ts"
 ---
+# [Stack] Rules
+
 [Stack-specific rules: error handling, imports, typing, testing patterns]
 ```
 
+Also generate a sensitive files guardrail:
+
+**File:** `.github/instructions/sensitive-files.instructions.md`
+```markdown
+---
+applyTo: "**/.env*,**/credentials*,**/secrets*,**/*.key,**/*.pem,**/appsettings.*.json"
+---
+# Sensitive Files — Do Not Edit
+
+- NEVER modify these files directly
+- NEVER write secrets, API keys, or passwords into any file
+- Move all secrets to environment variables
+```
+
 ### Step 4 — `.github/agents/` (agent definitions)
 
 Create one file per agent: `explorer.agent.md`, `architect.agent.md`, `coder.agent.md`, `tester.agent.md`, `reviewer.agent.md`.
 
 Each file:
-```markdown
+```yaml
 ---
-name: [Agent Name]
+name: [agent-name]
 description: "[One-line role description]"
+model: "Claude Sonnet 4"
+tools:
+  - read
+  - search
+---
+```
+
+Tool aliases per agent:
+- explorer/architect/reviewer → `["read", "search"]`
+- coder/tester → `["read", "edit", "search", "execute"]`
+
+Model per agent:
+- explorer/coder/tester → `"Claude Sonnet 4"`
+- architect/reviewer → `"Claude Opus 4"`
+
+Body includes: Role, Responsibilities, Approach, Output format.
+Body max: 30,000 characters.
+
+### Step 5 — `.github/skills/` (native skills)
+
+Copy and adapt each core skill:
+- `full-workflow/SKILL.md`
+- `hotfix/SKILL.md`
+- `explore-and-plan/SKILL.md`
+- `review-pr/SKILL.md`
+
+Frontmatter format:
+```yaml
+---
+name: "full-workflow"
+description: "Execute the complete development workflow: explore, implement, test, review."
+user-invocable: true
+---
+```
+
+### Step 6 — `.github/prompts/` (IDE fallback)
+
+For each skill, create a simplified prompt file as fallback:
+```yaml
+---
+description: "Complete development workflow: explore, implement, test, and review."
+agent: "agent"
+tools:
+  - read
+  - edit
+  - search
+  - execute
 ---
-# [Agent Name]
+# [Skill Name]
 
-**Role:** [role]
-**Responsibilities:** [responsibilities]
-**Approach:** [how it works]
-**Output:** [what it produces]
+[Simplified single-turn workflow steps]
 ```
 
 ---
@@ -130,8 +231,10 @@ crewkit-setup complete
 Generated:
   .ai/memory/                    (6 files)
   .github/copilot-instructions.md
-  .github/instructions/          (per-stack rules)
+  .github/instructions/          (per-stack rules + sensitive-files guardrail)
   .github/agents/                (5 agents)
+  .github/skills/                (4 native skills)
+  .github/prompts/               (4 IDE fallback prompts)
 
 Stack detected: [stack]
 Architecture:   [pattern]
@@ -139,4 +242,5 @@ Architecture:   [pattern]
 Next steps:
   1. Review .github/copilot-instructions.md — adjust hard rules if needed
   2. Open Copilot Chat — context is now active
+  3. Try: @coder implement [task] or /full-workflow [task]
 ```

package/src/cli.js

@@ -7,7 +7,7 @@ const HELP = `
 crewkit — Context engineering for AI-assisted development
 
 Commands:
-  install       Install crewkit globally (Claude Code, Copilot, Cursor)
+  install       Install crewkit globally (Claude Code, Copilot)
   update        Update to latest version (re-run install)
   add <name>    Add an optional skill or pack to the current project
   list          List all available skills and packs (core + add-ons)

package/src/detect.js

@@ -16,10 +16,5 @@ export function detectTools() {
     tools.push({ id: 'copilot', name: 'GitHub Copilot', dest, versionFile: join(dest, 'crewkit-setup.version') });
   }
 
-  if (existsSync(join(home, '.cursor'))) {
-    const dest = join(home, '.cursor');
-    tools.push({ id: 'cursor', name: 'Cursor', dest, versionFile: join(dest, 'crewkit-setup.version') });
-  }
-
   return tools;
 }

package/src/install.js

@@ -37,7 +37,7 @@ export async function install() {
     console.log(`
   No supported AI tools detected.
 
-  crewkit looks for: ~/.claude, ~/.copilot, ~/.cursor
+  crewkit looks for: ~/.claude, ~/.copilot
   Install one of these tools and re-run.
     `);
     process.exit(1);
@@ -56,10 +56,6 @@ export async function install() {
         installSingleFile(target, 'copilot-agent.md', 'crewkit-setup.md',
           'GitHub Copilot  ', 'Use @crewkit-setup in Copilot CLI or Chat', version);
         break;
-      case 'cursor':
-        installSingleFile(target, 'cursor-global.md', 'crewkit-setup.md',
-          'Cursor          ', 'Copy to .cursor/rules/ in your projects', version);
-        break;
     }
   }
 }

package/skill/adapters/cursor.md

@@ -1,218 +0,0 @@
-# Adapter: Cursor
-
-This adapter is executed during Phase 7, Step 10 of `/crewkit-setup`.
-You are the AI. Follow every instruction in this file to generate Cursor-compatible context files.
-
-**Input (two modes):**
-- **With Claude files:** If `CLAUDE.md` and `.claude/` exist (Claude Code was also a target), use them as primary source and reformat for Cursor.
-- **Standalone:** If Claude files do NOT exist, generate directly from `.crewkit/last-scan.md` (project profile) + `.ai/memory/` (architecture, conventions, commands). This is the primary source of truth.
-
-**Output:** Files under `.cursor/rules/` and `AGENTS.md` at the project root.
-
----
-
-## Rules for this adapter
-
-1. All generated files MUST be in **English**.
-2. Do NOT duplicate `.ai/memory/` — it is shared between all IDEs. No transformation needed.
-3. `model:` frontmatter from `.claude/agents/*.md` is Claude Code-only — strip it.
-4. Cursor has no equivalent for skills/prompts — skip `.claude/skills/` entirely.
-5. Create `.cursor/rules/` directory if it does not exist.
-
----
-
-## Step U1 — `.cursor/rules/project.md`
-
-**Source:** `CLAUDE.md` if it exists, otherwise `.crewkit/last-scan.md` + `.ai/memory/conventions.md`
-**Transformation:** If using CLAUDE.md, reformat for Cursor — add required frontmatter, remove agent/skill/hook sections. If generating from scan data, create the content directly using the project profile and conventions.
-
-**Required Cursor frontmatter:**
-```markdown
----
-description: "Project rules"
-alwaysApply: true
----
-```
-
-**Expected output format:**
-```markdown
----
-description: "Project rules"
-alwaysApply: true
----
-# [PROJECT NAME] — Project Rules
-
-## Overview
-[1-2 sentences from CLAUDE.md overview — what the project is, main stack]
-[Business domain: what it does, core entities, risk profile]
-
-**Stack:** [stacks]
-**Architecture:** [key patterns]
-
----
-
-## Hard rules (apply to every response)
-
-[Numbered list — copy from CLAUDE.md hard rules verbatim.]
-
-1. [Rule 1]
-2. [Rule 2]
-...
-
-Details for each rule → `.ai/memory/conventions.md`
-
----
-
-## Project Memory (`.ai/memory/`)
-
-Load context on demand:
-
-| File | When to load |
-|------|-------------|
-| `architecture.md` | Always — modules, layers, dependencies |
-| `conventions.md` | Always — naming, patterns, anti-patterns |
-| `commands.md` | When running build/test/deploy |
-| `testing.md` | When creating or running tests |
-| `lessons-{domain}.md` | When working on that domain |
-
----
-
-## Output Format
-
-Always return:
-- **Summary** — what was done
-- **Files changed** — list with brief description
-- **Tests** — pass/fail count (if tests were run)
-- **Risks / Next steps** — if any
-```
-
-**What to REMOVE from CLAUDE.md:**
-- `## Agent Discipline` section
-- `## Skills (slash commands)` section
-- `## Architect Decision Gate` section
-- `## Test Safety Loop` section
-
----
-
-## Step U2 — `.cursor/rules/*.md`
-
-**Source:** `.claude/rules/*.md` if they exist, otherwise generate directly from scan data (detected stacks + HIGH/MEDIUM confidence patterns).
-**Transformation:** If using Claude rules, convert frontmatter to Cursor format — keep glob patterns and rule content unchanged. If generating from scan data, create one rules file per detected stack with appropriate glob patterns and rules.
-
-Claude Code frontmatter format:
-```markdown
----
-description: "Node.js coding rules — applied when editing src/**/*.{js,ts}"
-globs: "src/**/*.{js,ts}"
----
-```
-
-Cursor frontmatter format:
-```markdown
----
-description: "Node.js coding rules"
-globs: "src/**/*.{js,ts}"
----
-```
-
-**Mapping:**
-- `globs:` → `globs:` (keep as-is — Cursor uses the same key)
-- `description:` → keep as-is, but shorten to the rule name without the "applied when editing..." suffix if present
-- Body: copy verbatim
-
-**File naming:** `.claude/rules/dotnet.md` → `.cursor/rules/dotnet.md`
-Keep the same filename — just place it under `.cursor/rules/`.
-
-**Example — source `.claude/rules/python.md`:**
-```markdown
----
-description: "Python coding rules — applied when editing **/*.py"
-globs: "**/*.py"
----
-
-# Python Rules
-
-- Use type hints on all function signatures
-- Validate input with Pydantic models at API boundaries
-```
-
-**Example — target `.cursor/rules/python.md`:**
-```markdown
----
-description: "Python coding rules"
-globs: "**/*.py"
----
-
-# Python Rules
-
-- Use type hints on all function signatures
-- Validate input with Pydantic models at API boundaries
-```
-
----
-
-## Step U3 — `AGENTS.md` (project root)
-
-**Source:** `.claude/agents/*.md` if they exist, otherwise generate directly from the agent templates at `~/.claude/skills/crewkit-setup/templates/agents/` (or from scan data if templates are unavailable).
-**Transformation:** Concatenate all agents into a single markdown file with `##` sections. Strip `model:` frontmatter from each. Remove the `<!-- crewkit:context-start -->...<!-- crewkit:context-end -->` block from each agent. Keep the `name:` and `description:` from frontmatter and all agent instructions. When generating from templates, inject project context from `.crewkit/last-scan.md` into each agent.
-
-**Output format:**
-```markdown
-# Agents
-
-This file describes the AI agents available in this project.
-Each agent has a specific role and scope. Invoke the appropriate agent for each task type.
-
----
-
-## [Agent Name]
-
-> [description from frontmatter]
-
-[agent body — full instructions, stripped of model: and crewkit context block]
-
----
-
-## [Agent Name 2]
-
-> [description from frontmatter]
-
-[agent body]
-
----
-```
-
-**Agent order:** explorer, architect, coder, tester, reviewer (same order as in `.claude/agents/`).
-
-**What to strip from each agent:**
-- `model:` frontmatter line
-- `name:` frontmatter line (becomes the `## heading` instead)
-- `description:` frontmatter line (becomes the `> blockquote` instead)
-- The entire `<!-- crewkit:context-start -->...<!-- crewkit:context-end -->` block (inclusive)
-- The YAML frontmatter delimiters (`---`) — the content moves to the `##` section body
-
-**File location:** `AGENTS.md` at the project root (not under `.cursor/`).
-
----
-
-## Step U4 — Skills
-
-**Source:** `.claude/skills/`
-**Action:** Skip entirely. Cursor has no equivalent concept for skills or prompts.
-
-Do NOT generate any file for this step. Log: "Cursor adapter: skills skipped (no Cursor equivalent)."
-
----
-
-## Completion Checklist — Cursor Adapter
-
-Before reporting done, verify each item:
-
-- [ ] `.cursor/rules/project.md` — exists, has `alwaysApply: true` frontmatter, contains hard rules, does NOT contain Agent Discipline or slash command sections
-- [ ] `.cursor/rules/` — one `.md` file per `.claude/rules/*.md` source file (plus `project.md`)
-- [ ] `.cursor/rules/*.md` — each has `globs:` frontmatter matching the source rule file
-- [ ] `AGENTS.md` — exists at project root, has `##` section for each of the 5 agents
-- [ ] `AGENTS.md` — no `model:` lines, no `crewkit:context-start` blocks, no YAML frontmatter delimiters
-- [ ] `.ai/memory/` — NOT duplicated under `.cursor/` (shared, no copy needed)
-- [ ] `.claude/skills/` — NOT copied (no Cursor equivalent, intentionally skipped)
-- [ ] No Portuguese in any generated file

package/skill/cursor-global.md

@@ -1,157 +0,0 @@
----
-description: "Context engineering setup — scan codebase and generate AI context files (.cursor/rules, AGENTS.md, .ai/memory)"
-alwaysApply: false
----
-
-# crewkit-setup
-
-Scan the current project and generate a complete context engineering setup tailored to this codebase.
-
-**You are the AI executing this.** Follow each phase in order. Do not ask the user questions. Scan, generate, report.
-
----
-
-## Phase 1 — Reconnaissance
-
-Scan the project to identify:
-- **Stack & frameworks**: languages, build tools, test frameworks, package managers
-- **Architecture**: monolith/microservices, key directories, entry points
-- **Existing AI context**: any `.cursor/`, `.github/`, `.ai/`, `CLAUDE.md`, `AGENTS.md` files
-
-Write findings to `.crewkit/scan-phase1-recon.md`.
-
----
-
-## Phase 2 — Codebase Mapping
-
-Read representative source files (3-5 per layer: models, services, controllers, tests). Identify:
-- Naming conventions, code style
-- Core domain concepts and entities
-- Test patterns and coverage approach
-- CI/CD and build commands
-
-Write findings to `.crewkit/scan-phase2-map.md`.
-
----
-
-## Phase 3 — Profile Compilation
-
-Synthesize phases 1-2 into a project profile:
-- Stack summary
-- Architecture summary
-- Key commands (build, test, lint)
-- Domain concepts
-- Hard rules to enforce
-
-Write profile to `.crewkit/last-scan.md`.
-
----
-
-## Phase 4 — File Generation
-
-Generate all files using the profile from `.crewkit/last-scan.md`.
-
-### Step 1 — `.ai/memory/`
-
-Create these files (all in English):
-- `architecture.md` — system design, key components, data flow
-- `conventions.md` — naming rules, patterns, anti-patterns
-- `commands.md` — build, test, lint, run commands with flags
-- `testing.md` — test strategy, frameworks, coverage approach
-- `lessons.md` — known gotchas, non-obvious behaviors
-- `state.md` — current sprint goal, recent changes, open issues
-
-### Step 2 — `.cursor/rules/project.md`
-
-```
----
-description: "Project rules"
-alwaysApply: true
----
-# [PROJECT NAME] — Project Rules
-
-## Overview
-[1-2 sentences: what the project is, main stack]
-
-**Stack:** [stacks]
-**Architecture:** [key patterns]
-
----
-
-## Hard rules
-
-[Numbered list of non-negotiable rules from the profile]
-
----
-
-## Memory
-
-Reference files in `.ai/memory/` for detailed context:
-- `.ai/memory/architecture.md` — system design
-- `.ai/memory/conventions.md` — coding conventions
-- `.ai/memory/commands.md` — build/test commands
-```
-
-### Step 3 — `.cursor/rules/` (per-stack rules)
-
-For each detected stack (e.g., `typescript.md`, `python.md`), generate a rules file with:
-```
----
-description: "[Stack] coding rules"
-globs: ["**/*.ts"] (or appropriate glob)
-alwaysApply: false
----
-[Stack-specific rules: error handling, imports, typing, testing patterns]
-```
-
-### Step 4 — `AGENTS.md`
-
-Create a single `AGENTS.md` at the project root with all agents as `##` sections:
-
-```markdown
-# Agents
-
-## Explorer
-Role: Codebase reconnaissance and research.
-[...]
-
-## Architect
-Role: System design and planning.
-[...]
-
-## Coder
-Role: Implementation.
-[...]
-
-## Tester
-Role: Test writing and validation.
-[...]
-
-## Reviewer
-Role: Code review and quality.
-[...]
-```
-
-Each agent section should include: Role, Responsibilities, Approach, Output format.
-
----
-
-## Phase 5 — Completion Report
-
-After all files are generated, report to the user:
-
-```
-crewkit-setup complete
-
-Generated:
-  .ai/memory/          (6 files)
-  .cursor/rules/       (project.md + per-stack rules)
-  AGENTS.md            (5 agents)
-
-Stack detected: [stack]
-Architecture:   [pattern]
-
-Next steps:
-  1. Review .cursor/rules/project.md — adjust hard rules if needed
-  2. Open Cursor and start coding — context is active
-```