Agent Engineering HandbookGitHub

Find a method for the failure you face

Each of the thirteen guides gives a concrete method, when to use it and what it can verify. Browse by source: the video, the Melee and Course Video Manager snapshots, or Boris Cherny's public work.

Browse the guides by source

Scroll across the thirteen guides and use the labels above them to identify each source.

  1. Guide 01: Convert recurring failures into permanent rules

    When your team keeps correcting the same mistake instead of checking for it automatically.

  2. Guide 02: One critical journey test beats broad shallow coverage

    When your product has a critical user journey that must never break silently.

  3. Guide 03: Make previews usable from the agent's environment

    When people or agents need to inspect changes across different worktrees or machines.

  4. Guide 04: Let the agent read and resolve CI failures

    When you keep copying CI errors from a web page into an agent prompt.

  5. Guide 05: Put project knowledge where the next task needs it

    When agents or newcomers keep missing project decisions, conventions or facts that are easy to overlook.

  6. Guide 06: Build small tool adapters for capability gaps

    When your agent needs an operation that its current tools do not support.

  7. Guide 07: Turn newcomer questions into improvements

    When teammates or agents keep getting stuck on the same first task.

  8. Guide 08: Keep contracts consistent across the application

    When agents create mismatched data shapes across storage, APIs and the user interface.

  9. Guide 09: Verification contracts

    When a finished command leaves you unsure whether the output meets its requirements.

  10. Guide 10: Codebase navigation and tooling

    When you need to navigate unfamiliar code and handle failures from external tools safely.

  11. Guide 11: One domain language and one transport when an agent calls your API

    When you add a command, route or agent operation to a codebase with a glossary and deployed API.

  12. Guide 12: Make pipeline artifacts reusable, resumable and honestly complete

    When slow, expensive work needs reusable outputs, recovery after crashes and checks for truncated artifacts.

  13. Guide 13: Validate the output through several useful views

    When ordinary fixtures miss failures in a generator, parser, migration, transformation or widely used library.

+ Theo's video

Eight guides adapted from the July 21, 2026 video. All quotations come from the video, including Boris's words as quoted by Theo. Investigation: Theo: T3 Code and Melee.

Guides 01–08

+ Melee for Mac

Two guides adapted from the verify script and the code map Theo keeps in his Melee for Mac repository. Investigation: Theo: T3 Code and Melee.

Guides 09–10

+ Matt Pocock's Course Video Manager

Two guides adapted from the checks and scripts in Matt Pocock's course-video-manager repository. Investigation: Matt Pocock: Course Video Manager.

Guides 11–12

+ Boris Cherny's public work

One guide adapted from json-schema-to-typescript CI checks: fixtures, built-artifact smoke tests, fuzzing, a real-world corpus and conformance checks. Investigation: Boris Cherny: public engineering work.

Guide 13