Skip to content
Aditya Karnam
AI researcher building the infrastructure layer for reliable agents.
Resume
Aditya Karnam · AI Research Notes
Free practical systems guide · Edition 1

Building a Coding Agent Harness

From model calls to a safe, observable, evaluable runtime. Build the runnable Rust core, then study how its boundaries map to Quecto in production.By Aditya Karnam and Arjun Jaggi. Chapters 2–5 develop a compact teaching harness you can run; Chapters 6–12 clearly mark production-runtime concerns as design studies.
Download the PDFOpen the accessible HTML edition ↗Get source + examples (.zip)
Cover of Building a Coding Agent Harness by Aditya Karnam and Arjun Jaggi
Pages75
Chapters14
ImplementationRust
LicenseMIT
Read Online

Preview the complete book

The PDF is embedded below. If your browser does not display it inline, use Download the PDF above or open the standalone HTML edition.
What You Will Build

A small, testable harness — one boundary at a time

The reference crate keeps the executable core deliberately small, then uses the production Quecto project to ground the wider runtime architecture.
Runnable coreModel transport, a bounded tool loop, typed tools, repository context, and execution policy, assembled and tested in the book's Rust crate.
Production concernsVerification, sessions, configuration, MCP, telemetry, and evaluation—explicitly labeled as design extensions where the compact crate does not implement them.
Quecto as evidenceA source-mapped tour of the larger open-source Rust system, including the production crates, features, and boundaries behind the examples.
For working engineersShort code excerpts, failure modes, invariants, build checkpoints, and exercises make the guide useful at the terminal, not just on the shelf.
Contents

Fourteen chapters, from API boundary to production system

  1. 01The Harness Is the System
  2. 02One Model Call, No Framework
  3. 03The Bounded Agent Loop
  4. 04Typed, Unsafe-Free Communication
  5. 05Execution Policy and Approval Gates
  6. 06Verification as a Completion Gate
  7. 07Context and Instruction Loading
  8. 08Sessions and State Management
  9. 09Profiles and Security Gates
  10. 10MCP: Model Context Protocol
  11. 11Observability and Telemetry
  12. 12Evaluation and Benchmarks
  13. 13The Complete Reference Harness
  14. 14Appendix: The Quecto Reference
The proof project

Quecto: the system behind the examples

The book is a general guide to harness design, with Rust as its implementation language and Quecto as a production reference—not a claim that the compact teaching crate duplicates every Quecto feature.
Explore Quecto on GitHub ↗Download the reference source package ↓
© 2026 Aditya Karnam. AI Researcher.
NowStackField NotesCurrent SystemsStatus