Agent skill · research science · aradotso
harness-engineering-cc-to-ai-coding
harness-engineering-cc-to-ai-coding — an agent skill for coding agents.
Why this skill is useful
Provides deep insights into Claude Code's architecture and best practices for AI coding that the AI wouldn't reliably generate on its own.
What it needs
Requires mdbook installed locally. About 6k tokens when loaded. Last updated 2026-07-09. 70 stars on the source repository.
What this skill does
bash Via Rust/Cargo cargo install mdbook Via Homebrew (macOS) brew install mdbook Via pre-built binary (Linux/macOS/Windows) Download from: https://github.com/rust-lang/mdBook/releases bash git clone https://github.com/ZhangHanDong/harness-engineering-from-cc-to-ai-coding.git cd harness-engineering-from-cc-to-ai-coding Build the book mdbook build book Serve locally with live reload mdbook serve book Default: http://localhost:3000 book/ ├── src/ # Markdown source files │ ├── SUMMARY.md # Table of contents / navigation │ ├── assets/ # Images (cover, diagrams) │ └── /.md # Chapter files └── book.toml # mdBook configuration Context Window ├── System Prompt (static, cached) ├── Tool Definitions (static, cached) ├── Conversation History (dynamic) │ ├── Full messages (recent) │ └── Compressed summaries (older) ← Auto-compression kicks in here └── Current Task Buffer Permission Levels (least → most privileged) ├── read-only — File reads, directory listing, search ├── write — File writes, creation, deletion ├── execute — Shell commands (bash tool) └── network — HTTP requests, external services [CACHED BOUNDARY] System Prompt Tool Definitions Common File Contents ← Extended cache prefix [DYNAMIC BOUNDARY] Conversation turns ← Not cached (changes each turn) Current user message Orchestrator Agent ├── Sub-agent A: "Search codebase for X" ├── Sub-agent B: "Write tests for Y" └── Sub-agent C: "Fix linting errors in Z" ↓ Results merged back into orchestrator context python class TokenBudgetManager: def init(self, maxtokens: int, compressionthreshold: float = 0.8): self.maxtokens = maxtokens self.compressionthreshold = compressionthreshold self.history: list[dict] = [] def shouldcompress(self, currenttokencount: int) -> bool: return currenttokencount / self.maxtokens > self.compressionthreshold def compresshistory(self, llmclient, history: list[dict]) -> list[dict]: """Summarize older turns to free token budget.""" if len(history) <= 4: return history # Keep at least 2 exchange pai …
How to use it
Reference it in AdaL, Claude Code, Cursor or any coding agent — nothing to install:
@skills aradotso/harness-engineering-cc-to-ai-coding