azclaude-copilot 0.1.8 → 0.2.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "azclaude-copilot",
-  "version": "0.1.8",
+  "version": "0.2.0",
   "description": "Autonomous product builder. Describe once, AZCLAUDE builds it across sessions — planning, implementing, testing, evolving, deploying. Zero human input.",
   "bin": {
     "azclaude": "./bin/cli.js",

package/templates/capabilities/evolution/generate.md

@@ -44,8 +44,16 @@ Sparse documentation is a signal, not a gap to fill with code.
 
 ### Step 3: Generate the File
 
-Write the capability file following the architecture rules:
-- ≤ 150 lines
+**If generating a skill/command**: read `capabilities/level-builders/level3-skills.md` first.
+Follow the Pushy Description Rule (10+ trigger scenarios), RECIPE pattern (steps not docs),
+and CE 2.0 Self-Correction layers. Skills go in `.claude/commands/{name}.md`.
+
+**If generating an agent**: read `capabilities/level-builders/level5-agents.md` first.
+Follow the 5-layer structure (Persona, Scope, Tools, Constraints, Domain Context).
+Use co-change analysis for scope boundaries. Agents go in `.claude/agents/cc-{name}.md`.
+
+**For all capability files** — follow these architecture rules:
+- ≤ 150 lines (overflow goes to references/ subdir)
 - YAML frontmatter required (name, description, tokens estimate)
 - One capability per file — if it's growing beyond 150 lines, split it
 - No file reads another file by default — explicit pointers only

package/templates/commands/evolve.md

@@ -152,23 +152,45 @@ If the file doesn't exist, create it with this header:
 
 ---
 
-## Step 7: Generate Project-Specific Skills and Agents from Evidence
+## Step 7: Generate or Update Project-Specific Skills and Agents
 
-### Skills from git evidence
+**Principle**: Never let Claude work without project-specific skills. Generate minimum viable skills early, then improve them as evidence accumulates.
+
+### 7a: Ensure minimum skills exist
+```bash
+ls .claude/commands/*.md 2>/dev/null | wc -l
+ls .claude/skills/*/SKILL.md 2>/dev/null | wc -l
+```
+If project has ZERO project-specific skills (only the 26 standard commands):
+1. Read `.claude/capabilities/level-builders/level3-skills.md`
+2. Analyze the project's stack, domain, and file structure
+3. Generate at least 2 skills immediately — even with zero git history
+4. Focus on: what file types exist → what workflow creates more of them
+
+### 7b: Generate new skills from git evidence
 ```bash
 git log --name-only --format="" --diff-filter=AM | sort | uniq -c | sort -rn | head -20
 ```
-If a file pattern appears 5+ times (e.g., `src/components/*.tsx`, `content/*.md`, `api/routes/*.py`):
+**Threshold: 2+ occurrences** (not 5 — skills should appear early, not late):
 1. Read `.claude/capabilities/level-builders/level3-skills.md`
 2. Create a skill that encodes the workflow for that pattern
 3. Save to `.claude/commands/{skill-name}.md`
 4. Only create if no existing skill covers this workflow
 
-### Agents from co-change clusters
+### 7c: Update existing skills from session learnings
+```bash
+ls .claude/memory/sessions/*.md 2>/dev/null | tail -3
+```
+If session files exist, scan for patterns that existing skills missed:
+- Repeated manual steps that should be in a skill → add them
+- Skill triggered but missing a step → update the skill
+- New conventions emerged → update skill's Rules section
+
+### 7d: Generate agents from co-change clusters
 ```bash
 git log --name-only --format="" --diff-filter=M | sort | uniq -c | sort -rn | head -30
 ```
-If 3+ files in the same directory change together across 3+ commits:
+If 3+ files in the same directory change together across 2+ commits:
 1. Read `.claude/capabilities/level-builders/level5-agents.md`
 2. Create an agent with all 5 layers in `.claude/agents/cc-{name}.md`
 3. Use co-change data for scope boundaries
@@ -176,9 +198,10 @@ If 3+ files in the same directory change together across 3+ commits:
 
 ### Check existing before creating
 ```bash
-ls .claude/commands/*.md .claude/agents/*.md 2>/dev/null
+ls .claude/commands/*.md .claude/agents/*.md .claude/skills/*/SKILL.md 2>/dev/null
 ```
 Skip creation if a skill/agent already covers the same workflow.
+Update existing ones if they're missing steps discovered in recent sessions.
 
 ---
 

package/templates/commands/setup.md

@@ -70,30 +70,41 @@ The orchestrator-init agent:
 
 ---
 
-## Step 5: Generate Project-Specific Skills
+## Step 5: Generate Project-Specific Skills (MANDATORY — minimum 2)
+
+**Rule**: Never leave a project without project-specific skills. Claude Code works significantly better with skills that encode project conventions. Generate at least 2 skills even on first run with zero git history.
 
 Read `.claude/capabilities/level-builders/level3-skills.md` for the skill design guide.
 
-Analyze the project's domain, stack, and recurring workflows:
-1. Check what file types and patterns exist (endpoints, pages, components, schemas, tests)
-2. Check git log for repeated commit patterns (what work is done most often)
-3. Identify 2-4 project-specific skills that would encode the most common workflows
+### How to decide which skills to generate
+1. **What file types exist?** → what workflow creates more of them
+2. **What domain is this?** → what domain-specific knowledge should be encoded
+3. **What stack?** → which stack-specific defaults apply
+4. Check git log for repeated commit patterns (if any history exists)
 
-For each skill:
-- Create `.claude/commands/{skill-name}.md` with pushy description (3+ trigger variants)
-- Follow RECIPE pattern (steps to do something, not documentation)
-- Include completion rule
-- ≤ 500 lines (references/ subdir for overflow)
+### Minimum: 2 project-specific skills
+Every project gets at least:
+1. A **creation skill** — how to add the main type of content/code in this project
+2. A **validation skill** — how to check that content/code follows project conventions
 
-**Stack-specific defaults** (generate if stack detected):
+### Stack-specific defaults (generate if stack detected):
 | Stack | Generate |
 |-------|---------|
 | Next.js / React | `new-page.md`, `new-component.md` |
-| Express / FastAPI | `new-endpoint.md` |
+| Express / FastAPI | `new-endpoint.md`, `validate-api.md` |
 | Any DB | `migrate.md` |
 | Any deploy config | `deploy.md` |
 | Markdown content | `new-content.md`, `validate-content.md` |
 | Exam / quiz | `new-exam.md`, `validate-exam.md` |
+| CLI tool | `new-command.md`, `validate-cli.md` |
+| Library/SDK | `new-module.md`, `validate-api-surface.md` |
+
+For each skill:
+- Create `.claude/commands/{skill-name}.md` with pushy description (10+ trigger variants)
+- Follow RECIPE pattern (steps to do something, not documentation)
+- Include project-specific conventions (file paths, naming, patterns from existing code)
+- Include completion rule
+- ≤ 500 lines (references/ subdir for overflow)
 
 **TaskUpdate → completed** for Create skills.