AgentSkin Suite
 
The complete token optimization stack for AI coding agents. Three complementary tools that eliminate the "Token Tax" across the full agent cycle — API responses, CLI output, and agent replies.
The Suite
| Component | What It Does | Savings | Creator | |-----------|-------------|---------|---------| | AgentSkin SSS | Semantic JSON pruning via MCP | 60–88% on rich APIs | Shawn Nichols Sr. | | Tokenjuice CLI | Rule-driven terminal output compaction | Up to 99.97% on large outputs | Vincent Koc (MIT) | | Caveman | Output compression via prompt engineering | 65% output reduction | Julius Brussee |
Quick Start
# 1. AgentSkin MCP server
npx -y agentskin@latest
# 2. Bash hook (transparent CLI optimization)
source .agents/hooks/bash-optimizer.sh
# 3. Caveman — already active via AGENTS.mdBy the Numbers
| Metric | Value | |--------|-------| | GitHub API savings | 88.3% (1,544 → 180 tokens) | | ls -laR directory listing | 99.97% (3.19M → 897 chars) | | Caveman output compression | 65% average | | Caveman memory compression | 46% smaller | | MCP tools | 7 across unified server | | Combined test suite | 4,695 tests, 274 files — 100% passing | | Pipeline throughput | 3,030 fixtures/sec (0.33ms avg) | | Net session savings | 17.1% (zero overhead via bash hook) |
Tools
AgentSkin Suite MCP (7 tools, unified)
fetch_optimized_data · skin_reasoning · classify_url · strip_ansi · reduce · estimate_tokens · apply_json_semantic
Caveman Skills (6 skills)
caveman · caveman-commit · caveman-review · caveman-stats · caveman-compress · cavecrew
Documentation
| Doc | Description | |-----|-------------| | OVERVIEW.md | Project overview and quick start | | ARCHITECTURE.md | Technical architecture and data flow | | STATUS.md | Phase completion and test coverage | | BENCHMARK.md | Live API compression benchmarks | | USAGE.md | Full usage guide — hook, caveman, MCP | | THREE-LAYER-RULES.md | Rule config override semantics | | AUTHORS.md | Credits and component details |
Security
- SSRF Protection: Blocks private IPv4/IPv6 ranges and cloud metadata services
- Rate Limiting: 60 req/min sliding window
- Input Validation: Zod schemas for all tool inputs
- Processing Timeout: 30-second limit
Credits
| Creator | Contribution | |---------|-------------| | Shawn Nichols Sr. (Nichols Transco LLC) | AgentSkin SSS protocol, MCP server, Suite integration | | Vincent Koc | Tokenjuice — MIT License | | Julius Brussee | Caveman — github.com/JuliusBrussee/caveman |
Website
agentskin.dev — Protocol specification, examples, FAQ, and whitepaper.
---
© 2026 Nichols Transco LLC.
