skill-create（技能创造）

skill-create

Guide users through creating new jaan.to skills with web research and best practices.

Context Files

• docs/extending/create-skill.md - Skill creation specification (v3.0.0)
• $JAAN_LEARN_DIR/jaan-to-skill-create.learn.md - Past lessons (loaded in Pre-Execution)
• $JAAN_TEMPLATES_DIR/jaan-to-skill-create.template.md - Generation templates
• $JAAN_CONTEXT_DIR/config.md - Current skill catalog
• ${CLAUDE_PLUGIN_ROOT}/docs/extending/language-protocol.md - Language resolution protocol

Input

Skill Idea: $ARGUMENTS

If provided, use as starting context. Otherwise, begin with identity questions.

Pre-Execution Protocol

MANDATORY — Read and execute ALL steps in: ${CLAUDE_PLUGIN_ROOT}/docs/extending/pre-execution-protocol.md Skill name: skill-create Execute: Step 0 (Init Guard) → A (Load Lessons) → B (Resolve Template) → C (Offer Template Seeding)

v3.0.0 Common Mistakes to Avoid (regardless of LEARN.md):

• ✗ Using hardcoded jaan-to/outputs/ instead of $JAAN_OUTPUTS_DIR
• ✗ Using hardcoded jaan-to/templates/ instead of $JAAN_TEMPLATES_DIR
• ✗ Using hardcoded jaan-to/learn/ instead of $JAAN_LEARN_DIR
• ✗ Using hardcoded jaan-to/context/ instead of $JAAN_CONTEXT_DIR
• ✗ Forgetting #anchor syntax when importing tech.md sections
• ✗ Using too-broad permissions like Write(jaan-to/**)
• ✗ Not validating with /jaan-to:skill-update before user testing

If the file does not exist, continue without it (but still avoid mistakes above).

Language Settings

Read and apply language protocol: ${CLAUDE_PLUGIN_ROOT}/docs/extending/language-protocol.md Override field for this skill: language_skill-create

PHASE 0: Duplicate Detection (Single Source of Truth)

Before any creation, check for existing skills:

1. Glob skills/*/SKILL.md to get all skills
2. For each skill, compare:

- Role + domain match - Purpose description similarity - Calculate overlap score (0-100%)

1. Decision tree:

- Exact match exists: "Skill '{name}' already does this. Use: /{command} [show example]" - >70% overlap: "'{name}' is similar ({n}% overlap). Update it instead? [update/new]" - If update: Invoke /jaan-to:skill-update {name} - If new: Continue with creation - <70% overlap: Continue with creation

1. Fast-track option for simple skills: "This seems straightforward. Create minimal skill directly? [y/wizard]" Skip wizard for:

- Single-purpose skills with obvious structure - Skills that wrap an existing command - Internal/utility skills with <50 lines expected

PHASE 1: Analysis (Interactive + Research)

Thinking Mode

ultrathink

Use extended reasoning for:

• Analyzing user requirements
• Planning skill structure
• Validating against specification
• Research synthesis

Questioning Guidelines

Ask up to 7 clarifying questions across Phase 1 if needed.

Skip questions when:

• Information already provided in user input
• Research answered the question
• Context files contain the answer
• Question is redundant with previous answer

Make questions smart:

• Reference user's input: "You said '{X}' — does that mean...?"
• Build on research: "Best practices suggest Y — does that fit your case?"
• Probe specifics: "What should happen when Z?"

Step 1: Basic Identity

Ask these questions one at a time:

After answers, validate and show:

"Skill name will be: {role}-{domain}-{action}" "Command: /jaan-to:{role}-{domain}-{action}" "Directory: skills/{role}-{domain}-{action}/"

Step 1.5: Check Project Configuration (v3.0.0)

Before proceeding with design, understand the project's configuration:

1. Check if configuration exists:

- Read jaan-to/config/settings.yaml (if exists) - Note any custom path configurations

1. Path customization check:

- Are default paths customized? - If settings.yaml has paths: section, note custom locations - Skills should use $JAAN_* env vars (automatically resolve to correct paths)

1. Learning strategy:

- Check settings.yaml for learning.strategy: "merge" or "override" - merge: Combine plugin + project lessons (default, recommended) - override: Use only project lessons (ignore plugin defaults)

1. Template customization:

- Check if custom templates exist for similar skills - Pattern: templates.{skill-name}.path: "./custom/path.md" - If project has custom templates, new skill should follow same pattern

Information helps generate skills that work correctly with the project's configuration.

"Configuration checked: [default paths / custom paths detected]"

Step 2: Web Research (Token-Optimized)

Use Task tool with Explore subagent to isolate research tokens:

Task prompt: "Research best practices for {domain} {action}:
1. Search '{domain} best practices {year}'
2. Search '{domain} report template'
3. Search 'how to {action} {domain}'
4. Search '{domain} checklist'

Return:
- 3-5 key best practices
- Suggested questions the skill should ask
- Suggested quality checks
- Suggested output sections
- Sources used"

Present research summary to user:

"Research findings for {domain}: Best Practices Found: 1. {practice1} 2. {practice2}... Suggested Questions for Skill: - {question1} - {question2}... Suggested Quality Checks: - {check1} - {check2}... Suggested Output Sections: - {section1} - {section2}..."

Step 3: Purpose & Triggers

Show research-based suggestions, then ask:

1. "What does this skill do? (1-2 sentences)"

- 2. "What phrases should auto-trigger this skill?" - [Suggested]: {action} {domain}, {domain} {action}, etc.

Step 4: Input & Output

1. "What input does the skill need?"

- Examples: file path, topic name, feature description - Will become $ARGUMENTS handling

1. "What files/outputs does it produce?"

- [Suggested from templates]: {format} file with {sections}

1. "What format? (markdown/json/both)"

- Determines template.md creation

Step 5: Questions, Quality & Done

Present research-based suggestions, let user accept/modify/add:

1. "What questions should the skill ask users?"

- [Pre-filled from research]: - {research_question1} - {research_question2} - "Add more or modify? [accept/edit]"

1. "What quality checks before writing?"

- [Pre-filled from research]: - {research_check1} - {research_check2} - "Add more or modify? [accept/edit]"

1. "What defines 'done' for this skill?"

- [Suggested]: - Output file written - Quality checks pass - User approved

HARD STOP - Human Review Check

Present complete skill structure:

SKILL SUMMARY
─────────────
Name: {role}-{domain}-{action}
Command: /{role}-{domain}-{action}
Logical: {role}-{domain}-{action}
Description: {description}

RESEARCH USED
─────────────
Sources: {source_count} web sources consulted
Best practices incorporated: {practice_count}

FILES TO CREATE
───────────────
□ skills/{name}/SKILL.md
□ skills/{name}/LEARN.md
□ skills/{name}/template.md (if needed)

WILL ALSO
─────────
□ Register in scripts/seeds/config.md
□ Create docs/skills/{role}/{name}.md (via /jaan-to:docs-create)
□ Commit to branch skill/{name}

"Create this skill? [y/n/edit]"

Do NOT proceed to Phase 2 without explicit approval.

PHASE 2: Generation (Write Phase)

Step 6: Create Git Branch

git checkout dev
git pull origin dev
git checkout -b skill/{role}-{domain}-{action}

Confirm: "Created branch skill/{name} from dev. All work on this branch."

Step 7: Generate SKILL.md

Use template from $JAAN_TEMPLATES_DIR/jaan-to-skill-create.template.md:

1. Fill YAML frontmatter:

- name: {name} - description: from Step 3 (must include "Use when" trigger phrase, no colons) - allowed-tools: based on needs from Step 5 - argument-hint: from Step 4 - license: PROPRIETARY - DO NOT add model: field (use inherited default)

1. Fill markdown body:

- Context Files from gathered info - Input handling from Step 4 - Phase 1 questions from Step 5 - HARD STOP section - Phase 2 generation steps - Quality checks from Step 5 - Definition of Done from Step 5

Step 8: Generate LEARN.md (Plugin Source)

Create skills/{name}/LEARN.md with research insights as initial lessons. Follow the standard LEARN.md structure from docs/extending/v3-compliance-reference.md section 12.7 (Better Questions, Edge Cases, Workflow, Common Mistakes). Seed each section with relevant research findings.

Step 9: Generate template.md (Plugin Source, if needed)

Based on output format from Step 4:

• Use researched report structure
• Include required metadata section
• Add placeholders for dynamic content

Step 10: Validate Against Specification

Reference: See ${CLAUDE_PLUGIN_ROOT}/docs/extending/skill-create-reference.md section "Specification Validation Checklist" for the full frontmatter, body, size, trust, and budget checks.

Validate against docs/extending/create-skill.md. If any check fails, fix before preview.

Step 11: Preview All Files

Show complete content of:

1. SKILL.md
2. LEARN.md
3. template.md (if created)

"Write these files? [y/n]"

Step 12: Write Files (v3.0.0-Compliant)

Reference: See docs/extending/v3-compliance-reference.md section "skill-create: v3.0.0 Best Practices for Writing Skills" for full details on Steps 12.1-12.9: - 12.1: Frontmatter (env var permissions, anti-patterns) - 12.2: Context Files section (standard env var references) - 12.3: Pre-Execution section (learning + language settings pattern) - 12.4: Template references (env vars, tech imports) - 12.5: Output paths (ID-based folder pattern, id-generator.sh, index-updater.sh) - 12.6: template.md variable syntax ({{field}}, {{env:VAR}}, {{config:key}}, {{import:path#section}}) - 12.7: LEARN.md seed structure (plugin-side, merge strategy) - 12.8: v3.0.0 validation checklist (compliance, output structure, tech stack, learning, quality) - 12.9: Automated path scan (hardcoded path detection and auto-fix)

Write to skills/{name}/SKILL.md, skills/{name}/LEARN.md, and skills/{name}/template.md (if needed), following all v3.0.0 patterns from the reference.

Confirm: "Skill files written to skills/{name}/ (v3.0.0-compliant)"

Step 13: Tech Stack Integration (Optional)

Ask: "Should this skill reference the project's tech stack?"

Reference: See docs/extending/v3-compliance-reference.md section "skill-create: Tech Stack Integration Reference" for full details on Steps 13.1-13.4 (identifying needs, updating SKILL.md, template imports, documentation).

Tech-aware: PRD generation, code generation, story writing, API docs. Tech-agnostic: Research, non-technical docs, roadmap planning.

"Tech integration: [enabled / not applicable]"

Step 14: Update Config Catalog

Edit scripts/seeds/config.md to add skill to Available Skills table:

| {role}-{domain}-{action} | `/{name}` | {short_description} |

Step 14.5: Update Team Roles Registry

Reference: See ${CLAUDE_PLUGIN_ROOT}/docs/extending/skill-create-reference.md section "Team Roles Registry Update Procedure" for the full role-matching and registry update steps.

If skill name matches a known role prefix, update skills/team-ship/roles.md accordingly. If roles.md does not exist, skip silently.

Step 15: Auto-Invoke Documentation

Run /jaan-to:docs-create to create:

• docs/skills/{role}/{name}.md

This ensures documentation is always created with the skill.

Step 16: Commit to Branch

Reference: See docs/extending/git-pr-workflow.md section "skill-create: Commit to Branch" for the full commit template.

Before staging, run:

bash scripts/prepare-skill-pr.sh

This regenerates + validates Codex skillpack artifacts and stages adapters/codex/skillpack/.

Then stage skills/{name}/, jaan-to/, and docs/skills/{role}/{name}.md. Commit with feat(skill) message including description and research source count.

PHASE 3: Testing & PR

Step 17: Validate v3.0.0 Compliance

Run /jaan-to:skill-update {skill-name} before user testing.

Reference: See docs/extending/v3-compliance-reference.md section "skill-create: v3.0.0 Post-Creation Validation" for full validation details.

Only proceed to user testing after validation passes.

Step 18: User Testing

"Please test the skill in a new session. Here's a copy-paste ready example:" `` /{name} "{example_input_based_on_skill_purpose}" ` For example, if the skill is pm-prd-write: ` /jaan-to:pm-prd-write "Add user authentication with OAuth support" `` "Did it work correctly? [y/n]"

If issues:

1. Help debug the problem
2. Make fixes
3. Commit fixes
4. Repeat testing

Step 19: Create Pull Request

When user confirms working:

"Create pull request to merge to dev? [y/n]"

Reference: See docs/extending/git-pr-workflow.md section "skill-create: Create Pull Request" for the full gh pr create template.

Push branch, create PR with skill summary, research used, files created, and an explicit line:

Codex skillpack sync: ✅ generated via scripts/prepare-skill-pr.sh

Show PR URL to user.

If no:

"Branch skill/{name} is ready. Merge manually when ready."

Step 20: Capture Feedback

"Any feedback on the skill creation process? [y/n]"

If yes:

• Run /jaan-to:learn-add skill-create "{feedback}"

Step 21: Auto-Invoke Roadmap Update

Run /jaan-to:pm-roadmap-update to sync the new skill with the roadmap.

This ensures the roadmap reflects the latest skill additions.

Skill Alignment

• Two-phase workflow with HARD STOP for human approval
• Single source of truth (no duplication)
• Plugin-internal automation
• Maintains human control over changes

Definition of Done

• Duplicate check completed
• Web research performed
• All skill files created (SKILL.md, LEARN.md, template.md)
• Passes specification validation
• scripts/validate-skills.sh passes (description budget + body line cap)
• Registered in context/config.md
• Documentation created via /jaan-to:docs-create
• User tested and confirmed working
• PR created (or branch ready for manual merge)
• Roadmap synced via /jaan-to:pm-roadmap-update
• User approved final result