What
A requirement specification for a complete capability.
A CRD describes the smallest complete, contextually meaningful ability that produces an outcome—not merely a feature, endpoint, component, or function.
Capability Documentation · CRD framework
A Capability Requirements Document for every complete capability.
Capability Documentation is the open framework. A capability may belong to a product, service, system, agent, or reusable open inventory; its Capability Requirements Document (CRD) makes purpose, boundaries, requirements, interaction contracts, and realizations understandable before that meaning fragments across UX, code, APIs, agents, and operations.
A Capability Requirements Document specifies one complete, contextually meaningful ability independently of its current implementation. It gives people, AI agents, applications, and developers one durable source for what the capability means and must keep true.
Product / Service / Agent
Capability Inventory
CRD per Capability MLE
Interaction Contracts
Realizations
UI / API / MCP / Skill / Code
For whom
People, agents, and software.
The framework gives product, domain, UX, and engineering people—as well as AI agents and applications—a shared language for understanding and using capabilities.
Why
Keep meaning from being lost.
It separates purpose, rules, defaults, examples, implementation choices, operational constraints, and unknowns so they are not silently confused or discarded.
How
Connect requirements to execution.
Define a Capability MLE, its Interaction Contract MLEs, and one or more realizations—then trace the CRD to the relevant UX, APIs, tools, code, tests, and operations.
Use with your AI agent
Give your agent the CRD context once.
Open the dedicated CRD prompt when an agent cannot install the CRD Author skill. It provides the canonical source order, vocabulary, authority rules, and transformation safeguards without overwhelming the homepage.
The agent is instructed to preserve unknowns, distinguish facts from inference, and not change a live service without explicit approval.
For agents that support installable skills: use the CRD Author skill. It provides Define and Extract workflows grounded in the same technology-agnostic model.
CRD agent resources
Use the complete prompt or installable skill.
The prompt is available as a focused HTML page, Markdown, and plain text. Agents that support skills can read the published SKILL.md directly.
Core model
Three layers, without implementation lock-in.
Capability MLE
The smallest complete, contextually meaningful ability that produces a meaningful outcome.
Interaction Contract MLE
The smallest contextually meaningful executable behaviour: actor, command, state, policies, transition, result, and effects.
Realization
One or more software, agent, or combined implementations of the capability or its contracts.
Origins
Why Minimum Logical Element matters.
MLE is the scope discipline beneath CRD: the smallest bundled unit that still retains logical sense and context across business, design, and development. It prevents both vague feature buckets and isolated technical fragments.
A Capability MLE is the complete-ability form of that idea; a CRD is its durable requirement specification.
Why it exists
Shared meaning across product, software, and agents.
CRD keeps general capability purpose separate from organization-specific rationale and implementation. It makes rules, defaults, examples, operational constraints, inferences, and unknowns explicit—so current code is not mistaken for the fundamental requirement.
Learning resource
Why CRD now?
As software moves from fixed applications toward capabilities realized by software, agents, or combinations of both, implementation is becoming increasingly fluid. CRD provides a stable semantic layer for what a system can meaningfully do, independently of how that capability is currently implemented.
PRDs, user stories, Storybook, APIs, MCP, skills, and code documentation remain useful. CRD provides the capability-level context that connects them.
Start here
Use the smallest useful form.
What CRD is, why it matters, and how to create or use one — in that order.
Learn CRD
Understand why capability-centered documentation matters, where MLE comes from, and how the pieces relate.
Start learning →CRD Specification
Required semantic core, optional extensions, and conformance rules.
Read →CRD template
Document a reusable capability without creating a mini-PRD.
Use →Worked examples
Illustrative quote-request and payment-reconciliation examples, GitHub's real pull-request pair, and an agent-primary scheduling example.
Explore →Advanced material
Once you have the basic model.
Capability Inventory
Catalogue available or planned Capability MLEs without requiring a product or PRD above them.
Browse →Source Context Reference
Preserve product-wide and cross-cutting source context that no single capability should own.
Use →Agent instructions
Transform plans, code, APIs, MCP, and existing material without inventing facts.
Apply →Working with CRDs
Use precedence, safe reuse, skills, tools, Related MLEs by Dimension, and audience projections without creating divergent requirements.
Guide →Media resources
Read, watch, or listen to source material on semantic blueprints, agent-ready documentation, and capability requirements.
Open →CRD Author skill
Give an installable AI-agent skill the CRD workflow for defining and extracting capabilities.
Open on GitHub →CRD vs PRD
Compare capability requirements with product requirements, feature specifications, user stories, APIs, skills, and tools.
Compare →CRD glossary
Use canonical definitions for the framework’s controlled vocabulary.
Browse terms →CRD prompt for AI agents
Give an agent the source order, vocabulary, authority model, and safe transformation workflow.
Open prompt →Agent-readable semantics
Make uncertainty and intent visible.
- Capability purpose
- Rationale / intent
- Rule / invariant
- Recommended default
- Example
- Implementation choice
- Operational constraint
- Unknown / unresolved