For AI Agents

Overview

Most of this site talks about C-DAD from a practitioner's chair: the person running cdad check, reading the report, deciding what to contract next. But the toolkit doesn't stop at a CLI a human types into. It ships six role prompts, plain markdown files that work as Claude Projects instructions, Cursor rules, or a Codex system prompt, each one written to do a specific job in the contract-driven workflow. They aren't generic "act as a software engineer" prompts. Every one of them is grounded in the toolkit's actual schemas, its worked examples, its CLI reference, and its own generated artifacts, so an agent running one of these prompts is reasoning from the same source of truth a practitioner would be reading.

The six roles split cleanly by what they touch. Five of them, the contract author, the contract reviewer, the legibility auditor, the extraction agent, and the contract maintenance agent, operate on contracts and audits rather than on application code. The sixth, the contract-aware coding agent, is the one that actually edits files, and it treats every contract it finds as the boundary of what it's allowed to change. None of them work in isolation. They compose into the same checkto roadmap to init to validateto graph loop the rest of the toolkit runs on, so picking the right prompt for the moment is mostly a matter of knowing where in that loop you currently stand.

Pages in this area

• The Contract-Aware Coding Agent

  The day-to-day implementation agent. It checks the relevant contract before touching code and never widens behavior without updating that contract first.

• The Contract Author

  Turns business intent into a complete contract.yaml, asking for capability ID, intent, inputs and outputs, non-goals, and constraint history, never implementation detail.

• The Contract Reviewer

  Reviews a contract in a fixed order: missing fields, implementation language disguised as intent, unsupported claims, missing constraints, then state mismatches.

• The Legibility Auditor

  Runs the book's four-question audit verbatim, starting from the diagnostic question, and ranks capabilities by how urgently each one needs extraction.

• The Extraction Agent

  Turns cdad check findings into a prioritized extraction sequence, weighing the legibility signal against business criticality and agent touchpoint frequency.

• The Contract Maintenance Agent

  Finds drift between contract.yaml, the code, and its generated json and markdown siblings, then recommends the exact semver bump to fix it.