Harness Engineering Series
Two books unpacking the design philosophies behind Claude Code & Codex—constraints, query loops, context governance, and multi-agent verification.
- Source repo
- wquguru/harness-books
- Stars
- ★ 3.2k
- Last updated
- 6mo ago
- Primary language
- Python
- FA score
- 26/100 · Major gaps
At a glance
- How it runs
- Works with
- Platform-specificCodex · Claude Code
- Cost
- Free, no paid service needed
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- An engineering lead evaluating how to integrate an AI coding agent into a team's workflow and needing a structured framework for governance.
- Not a fit if
- Teams wanting to integrate it into a workflow—it's reference reading, not a tool
- Chinese-only readers—the Chinese translation may lag the English version
- Teams on agent frameworks other than Claude Code and Codex—principles may not map
- Source review
- 26/100 · Major gaps 8 safety controls not found
What does this agent do, and when should you use it?
This repository hosts two books on harness engineering—the discipline of keeping code-writing models bounded, continuous, and accountable when placed inside terminals, repos, permission systems, and team workflows. Book 1 uses Claude Code as a case study to explore runtime structure: the control plane, query loop, tool permissions, context governance, recovery paths, multi-agent verification, and team rules. Book 2 compares the harness design of Claude Code and Codex, examining where each places order—one starts from runtime discipline, the other from a structured control layer. The books are not line-by-line code walkthroughs but focus on how a harness organizes constraints and execution. They are available free to read online and as PDF downloads.
The repository provides the content of two e-books (English and Chinese versions) with online readers and downloadable PDFs. The books explain the engineering rationale behind components like query loops, permission systems, context budgeting, error recovery, and multi-agent verification in AI coding tools. They offer checklists and diagrams to help teams translate principles into actionable constraints. Additionally, the repo includes build tooling (Python and Honkit scripts) to regenerate the sites locally.
- An engineering lead evaluating how to integrate an AI coding agent into a team's workflow and needing a structured framework for governance.
- A developer who wants to understand the internal control plane of Claude Code beyond surface-level prompt engineering.
- A technical decision-maker choosing between Claude Code and Codex based on architectural philosophy rather than feature checklists.
- A platform engineer building a custom harness for an AI agent and looking for design principles and source maps to study.
- An educator or writer preparing material on AI agent design and seeking a rigorous, well-organized reference.
How do you install or deploy this agent?
No installation required—read online. To build locally, need Python 3 and Honkit. Clone the repo and run the build scripts as described in the README.
How do you use this agent?
Visit https://harness-books.agentway.dev/en/ to read online or download PDFs. To build locally, execute python3 tools/book-kit/build_honkit.py book1-claude-code and python3 tools/book-kit/build_honkit.py book1-claude-code --locale en (and similarly for book2), then python3 tools/book-kit/build_pages_site.py; output goes to dist/.
What are this agent's strengths and limitations?
- Deep, structured analysis of harness design, not just a list of features—helps readers reason about system boundaries.
- Free online access and PDF downloads, plus a Chinese translation is included.
- Practical appendices with checklists and source maps that let readers tie concepts to real code files.
- Not a software tool, so it cannot be integrated into a workflow—it's reference material only.
- Chinese translation may lag behind the English version; check the English for the latest.
- Focuses on Claude Code and Codex; the principles may not map perfectly to other agent frameworks.
How does this agent compare with similar options?
The second book directly compares Claude Code and Codex, giving readers a structured way to evaluate both systems for their own needs.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Harness Engineering Series This agent | 26 · Major gaps | Web appFree | ★ 3.2k | 6mo ago | Python | Codex · Claude Code |
| Learn Harness Engineering | 32 · Major gaps | CLIFree + model costs | ★ 21k | 18d ago | TypeScript | Codex · Claude Code |
| AIWG | Insufficient evidence | CLIFree + model costs | ★ 224 | today | TypeScript | Codex · Claude Code |
| Better Harness | 85 · Good | Agent plugin / skillFree | ★ 2.4k | 6d ago | JavaScript | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows this is a documentation/books repository with no executable code or agent behavior. Therefore, all trust-related criteria (least privilege, user confirmation, data flow transparency, sensitive data handling, dependency security, external effects, rollback, source attribution) are not applicable and scored 0. No evidence of malicious behavior, but no trust mechanisms either.
Self-consistency: Book structure is clear, chapters and appendices cross-reference each other, content is coherent, score 2. Dependency availability: Build dependencies (e.g., pandoc, texlive) are listed in CI, but no version pinning or integrity checks, score 1. Failure messages: No documentation of failure handling or error messages, score 0.
Audience and scenarios: README clearly identifies target readers and reading paths, score 2. Capability boundaries: Book content explicitly states its scope (not line-by-line source analysis), but no explicit boundary statement, score 1. Trigger precision: Not applicable, score 0. Environment fit: Local build instructions provided, but no compatibility notes for different OS/environments, score 1.
Information architecture: Table of contents and chapter organization are good, score 2. Install notes: Local build steps provided, but no detailed dependency installation instructions, score 1. Naming stability: File and chapter naming consistent, but no version stability statement, score 1. Examples and FAQ: Examples exist (e.g., build commands), but no FAQ, score 1. Known limitations: Not mentioned, score 0. License: No license information, score 0. Versioning and changelog: Not provided, score 0. Maintenance responsibility: No clear maintainer or contribution guidelines, score 1.
Output usability: Books are available in HTML and PDF, easy to read, score 2. Marginal value: Provides in-depth content on harness engineering, unique value, score 2. Cost-benefit: Free, but no maintenance cost or update frequency, score 1.
Claim traceability: Books claim to be based on source code, but no specific citations, score 2. Cross-source corroboration: No external sources or verification, score 1. Fact-inference separation: Content is mostly analysis and inference, but not clearly separated, score 1.
- Not found in source: least-privilege scopingGrant only what the task needs: a dedicated account or read-only token, scoped to specific directories and repos.
- Not found in source: confirmation before actingTurn on (or add) a confirmation step before it acts, and try it in a sandbox or test environment before real data.
- Not found in source: data-flow disclosureWatch which external services it contacts (proxy or firewall logs) and keep sensitive data out until you know where it goes.
- Not found in source: sensitive-data handlingUse dedicated, low-privilege, revocable API keys — never production credentials — and keep secrets out of logs.
- Not found in source: dependency securityPin versions and run a dependency audit (npm audit, pip-audit) before installing; prefer running it in a container.
- Not found in source: disclosed external effectsEstablish which external systems it writes to, sends to or changes, and verify with test accounts or repos before production.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- Not found in source: verifiable attributionInstall from the official repo or registry and check the publisher and URL to avoid look-alike packages.
- This repository is documentation/books, no executable code, so trust criteria are not applicable.
- No license information provided; confirm before use.
- No versioning or changelog; content may change over time.
- Book content is based on author analysis, no external verification.