---
paths: .claude/agents/**/*.md, .claude/commands/**/*.md
---
# Claude Extensions Development Rules
Quick reference for sub-agents and slash commands.
## Sub-Agent Frontmatter
```yaml
---
name: kebab-case-name
description: When to use. Add "Use PROACTIVELY" for auto-activation.
tools: Read, Write, Bash(pytest:*) # Minimal set
model: sonnet # sonnet | opus | haiku
---
```
## Slash Command Frontmatter
```yaml
---
description: One-line (shown in menu)
allowed-tools: Read, mcp__ib-sec-mcp__*
argument-hint: [symbol] [--flag]
---
```
## Model Selection
| Model | When to Use | Token Cost |
|-------|-------------|------------|
| `haiku` | Simple tasks, fast response | Low |
| `sonnet` | Balanced, default choice | Medium |
| `opus` | Complex analysis, implementation | High |
## Tool Patterns
```yaml
# Wildcard for MCP tools
tools: mcp__ib-sec-mcp__*
# Specific bash commands
tools: Bash(pytest:*), Bash(python:*)
# Read-only for analysis
tools: Read, Grep, Glob
# Full development
tools: Read, Write, Edit, Bash(python:*), Bash(pytest:*)
```
## Common Mistakes
- Missing `model:` field (defaults may not be optimal)
- Over-permissive tools (security risk)
- Missing "Use PROACTIVELY" for auto-activation
- No error handling section in commands
- Overly long descriptions (hard to scan)
## Sub-Agent Testing
- [ ] Explicit invocation works
- [ ] Auto-activation triggers correctly (if PROACTIVELY)
- [ ] Tools are sufficient but minimal
- [ ] Output format matches expectations
## Slash Command Testing
- [ ] Appears in command menu (`/` to see)
- [ ] Arguments parsed correctly (`$1`, `$2`, `$ARGUMENTS`)
- [ ] Default values work
- [ ] Error messages are helpful
See `.claude/SUB_AGENTS.md` and `.claude/SLASH_COMMANDS.md` for detailed guides.
