Features: - Multi-platform bot (Slack, Telegram) - Memory system with SQLite FTS - Tool use capabilities (file ops, commands) - Scheduled tasks system - Dynamic model switching (/sonnet, /haiku) - Prompt caching for cost optimization Optimizations: - Default to Haiku 4.5 (12x cheaper) - Reduced context: 3 messages, 2 memory results - Optimized SOUL.md (48% smaller) - Automatic caching when using Sonnet (90% savings) Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
2.1 KiB
2.1 KiB
Local Skills for Ajarbot
This project uses local-only skills for security - no public registries, no external dependencies.
Available Skills
/adapter-dev - Adapter Development Helper
Helps create, debug, and update messaging platform adapters.
Usage:
/adapter-dev create WhatsApp adapter
/adapter-dev debug Slack connection issues
/adapter-dev add file upload support to Telegram
Location: .claude/skills/adapter-dev/
Creating Your Own Skills
1. Create skill directory
mkdir -p .claude/skills/my-skill
2. Create SKILL.md
---
name: my-skill
description: What this skill does
user-invocable: true
disable-model-invocation: true
allowed-tools: Read, Grep
context: fork
---
Instructions for Claude when this skill runs...
3. Use the skill
/my-skill
/my-skill with arguments
Security Configuration
Restrict skill permissions in .claude/settings.json:
{
"permissions": {
"allow": [
"Skill(adapter-dev)",
"Skill(my-skill)"
]
}
}
Per-skill tool restrictions in SKILL.md frontmatter:
allowed-tools: Read, Grep, Glob
This prevents skills from using Bash, Edit, Write, etc. unless explicitly allowed.
Why Local Skills?
✅ No supply chain attacks - All code is in your repo ✅ Version controlled - Review skills in PRs ✅ Team-wide consistency - Everyone uses same skills ✅ Fully auditable - All code is visible ✅ Offline capable - No registry lookups
Skill Arguments
Pass arguments to skills:
/adapter-dev create Discord adapter
In SKILL.md, access with:
$ARGUMENTS- All arguments as string$0,$1,$2- Individual arguments
Best Practices
- Use
disable-model-invocation: truefor security-sensitive skills - Limit
allowed-toolsto only what's needed - Use
context: forkto isolate skill execution - Document in examples/ directory
- Review all skills before committing