---

name: suggesting-tooling description: | Analyzes a codebase and suggests custom Claude Code skills and subagents to improve development workflow. Use when user wants tooling suggestions, automation ideas, or asks "what skills would help this project".

Suggesting Tooling

Analyzes codebases to suggest custom skills and agents for workflow automation.

Quick Start

1. Scan codebase for patterns (languages, frameworks, workflows)
2. Match patterns to skill/agent templates
3. Present suggestions with rationale
4. Generate approved tooling using creating-skills/creating-agents

Workflow

CopyProgress:
- [ ] Analyze codebase patterns
- [ ] Match to tooling templates
- [ ] Present suggestions
- [ ] Generate approved items

Step 1: Analyze Codebase

Run lightweight analysis:

Copy# Language detection
find . -type f -name "*.ts" -o -name "*.js" -o -name "*.py" | head -100

# Framework detection
ls package.json pyproject.toml Cargo.toml go.mod 2>/dev/null

# Workflow detection
ls .github/workflows/ .gitlab-ci.yml Dockerfile 2>/dev/null

# Existing tooling
ls .claude/skills/ .claude/agents/ 2>/dev/null

Collect:

• Primary language(s)
• Framework(s) in use
• Build/test tooling
• CI/CD setup
• Existing .claude/ configuration

Step 2: Pattern Matching

Match detected patterns to suggestions:

See patterns/skills.md and patterns/agents.md for complete mappings.

Step 3: Present Suggestions

Format suggestions for user review:

Copy## Tooling Suggestions for {project}

Based on analysis of your codebase:
- Language: {detected}
- Framework: {detected}
- Existing tooling: {count} skills, {count} agents

### Recommended Skills

| # | Skill | Why | Priority |
|---|-------|-----|----------|
| 1 | {name} | {rationale} | P1 |
| 2 | {name} | {rationale} | P2 |

### Recommended Agents

| # | Agent | Why | Priority |
|---|-------|-----|----------|
| 1 | {name} | {rationale} | P1 |

Then ask user which to generate.

Step 4: Generate Tooling

For each approved item:

Skills - Invoke creating-skills:

CopyUse the creating-skills skill to create a {name} skill.

Purpose: {rationale}
Detected context:
- Framework: {framework}
- Test runner: {runner}
- Patterns: {patterns}

Generate a focused skill for this project.

Agents - Invoke creating-agents:

CopyUse the creating-agents skill to create a {name} agent.

Purpose: {rationale}
Suggested tools: {tools}
Detected context:
- Project type: {type}
- Workflows: {workflows}

Generate a focused agent for this project.

Pattern Categories

Testing

• Signals: jest.config, pytest.ini, mocha, test/ directory
• Suggest: testing-code skill
• Priority: P1 if tests exist but no skill

API Development

• Signals: Express routes, FastAPI, REST patterns
• Suggest: api-testing agent
• Priority: P1 if API-heavy project

Database

• Signals: Prisma, TypeORM, migrations/
• Suggest: db-migrations skill
• Priority: P2

DevOps

• Signals: Dockerfile, docker-compose, CI configs
• Suggest: deploying-code skill
• Priority: P2 if no deployment automation

Code Quality

• Signals: ESLint, Prettier, pre-commit
• Suggest: linting-code skill
• Priority: P3

Security

• Signals: Auth middleware, JWT, OAuth
• Suggest: security-review agent
• Priority: P1 for auth-heavy projects

See reference.md for detailed pattern definitions.

Gap Analysis

Compare detected needs against existing .claude/ configuration:

CopyDetected workflows:     Existing tooling:
- Testing (Jest)        - (none)
- API (Express)         - (none)
- CI (GitHub Actions)   - (none)

Gaps: testing, api-testing, deployment

Only suggest tooling that fills gaps.

Output Format

After generation, report:

Copy## Tooling Created

| Type | Name | Location |
|------|------|----------|
| Skill | testing-code | .claude/skills/testing-code/ |
| Agent | code-reviewer | .claude/agents/code-reviewer.md |

### Restart Required

New skills and agents require a Claude restart to be available.

To continue where you left off:
\`\`\`bash
claude --continue
\`\`\`

### Next Steps

1. Restart Claude to load new tooling
2. Run `claude --continue` to resume
3. Review generated tooling in .claude/
4. Test skills with sample workflows

Limits

• Maximum 5 suggestions per category
• Require explicit approval before generating
• Skip suggestions for existing tooling
• Prioritize by impact (P1 first)