Reduce Playwright MCP snapshot token usage by ~90% using Haiku as a summarization layer
npx claudepluginhub treesoop/claude-native-plugin --plugin playwright-optimizerPlugins that save tokens, time, and money for Claude Code users.
/plugin marketplace add treesoop/claude-native-plugin
/plugin install playwright-optimizer
| Plugin | Description | Savings |
|---|---|---|
| playwright-optimizer | Summarize Playwright MCP snapshots with Haiku before they reach Opus | ~90% token reduction |
| task-tracker | Track Claude Code usage (tokens, time, projects) per team | Team analytics |
These plugins use Claude Code's native hook system to optimize expensive operations. No API key needed — uses claude CLI with your existing Claude Code subscription auth.
Playwright MCP sends massive YAML snapshots (10K-60K+ chars) for every browser interaction. This plugin intercepts them via PostToolUse hook, summarizes with Claude Haiku via claude -p --model haiku, and passes a compact summary to your primary model.
Playwright MCP → 60,000 char YAML snapshot
→ PostToolUse hook triggers
→ claude -p --model haiku (internal auth, no API key)
→ ~4,000 char summary (all ref= values preserved)
→ Opus receives compact summary only
Tested results:
| Metric | Before | After | Change |
|---|---|---|---|
browser_navigate | 37,983 chars | 1,922 chars | -94% |
browser_snapshot | 37,897 chars | 1,435 chars | -96% |
claude) installed and authenticatedjq installed (pre-installed on most systems)MIT
Ultra-compressed communication mode. Cuts ~75% of tokens while keeping full technical accuracy by speaking like a caveman.
Intelligent prompt optimization using skill-based architecture. Enriches vague prompts with research-based clarifying questions before Claude Code executes them
Qiushi Skill: methodology skills for AI agents guided by seeking truth from facts, with Claude Code, Cursor, OpenClaw, Codex, OpenCode, and Hermes guidance.