v0.1.0
Initial Release
A self-demonstrating Claude Code plugin that encodes heuristics for plugin design decisions -- skill vs. agent vs. command vs. hook.
What's Included
| Component | Purpose |
|---|---|
skills/plugin-design-advisor/ |
Core heuristics for choosing the right plugin mechanism |
skills/decision-explorer/ |
Interactive decision tree for learners |
agents/architecture-validator.md |
Audits existing plugin structure for misclassifications |
agents/requirements-analyzer.md |
Classifies task characteristics |
agents/constraint-extractor.md |
Identifies parallelism/isolation/tiering needs |
commands/design.md |
Multi-phase design workflow with parallel analysis |
commands/validate.md |
Run architecture-validator against a plugin directory |
hooks/hooks.json |
PreToolUse guard on Write/Edit to agents/ and skills/ |
Anti-Patterns Detected
- Agent-as-skill -- read-only agent with no isolation benefit
- Skill-as-agent -- domain knowledge buried in an agent prompt
- Procedure-as-skill -- procedural workflow stuffed into a skill
- Command-without-hook -- system-event workflow as explicit command
- Hook-without-suppression -- hook fires on every event without checking session state
- MCP-for-no-reason -- MCP server added without external service dependency
Implementation Phases
- Phase 1: Core skill + architecture-validator agent
- Phase 2: Design command + parallel analyzer agents with DECISIONS.md suppression
- Phase 3: plugin-file-guard PreToolUse hook with path filter and session dedup
- Phase 4: decision-explorer interactive skill with adaptive branching