Stop Building the Wrong Thing: The CRISP Framework Ships Production-Ready CLAUDE.md Files

Most AI implementations fail before a single line of code is written. Not because the model was wrong, but because nobody defined the problem. The newly open-sourced CRISP framework (Client-Ready Implementation Spec Process) directly addresses this by providing a structured 5-phase methodology that takes you from a vague client request to a fully specced, Claude Code-ready implementation brief.

What It Does — A Structured Pipeline from Brief to Build

CRISP is a BA/PM framework built specifically for AI implementors using tools like Claude Code. It's designed for AI agency owners, solo builders, consultants, and product managers who need a repeatable discovery-to-spec process. The framework's core insight: "Every AI project failure I've seen had the same root cause. Builder gets a brief → skips straight to the stack → builds something technically impressive → delivers it → client says 'this isn't what I meant.'"

CRISP fills that gap with explicit data pipelines where every document has an upstream source. Nothing is invented twice. Every deliverable is derived from prior phase outputs—not created from scratch.

Setup — Drop the /skills Folder and Start Templating

Getting started is straightforward:

1. Clone the repository: git clone https://github.com/radekamirko/C.R.I.S.P.git
2. Drop the /skills folder into your Claude Code project root
3. Use the /templates directory as your deliverable starting points
4. Fill outputs into your project's docs/ directory

The framework includes templates for:

• Problem Definition Documents
• Stakeholder Impact Maps
• Technical Stack Tables with version pinning
• AI Implementation Specs
• Risk Registers

When To Use It — Before Claude Code Touches a Sprint

The golden rule of CRISP: No spec = no build. Claude Code never touches a sprint without a locked AI Spec. Every open question in the spec must be resolved before the sprint starts. A spec with unanswered questions is a wishlist, not a brief.

This is particularly valuable for:

• Client work where requirements are vague or stakeholders have conflicting priorities
• Internal AI products where different departments have unstated goals
• Complex integrations where technical decisions have business implications

The CLAUDE.md Payoff — Security and Version Control Built-In

One of CRISP's most actionable outputs is the production-ready CLAUDE.md file it generates. This isn't a generic template—it's a project-specific configuration that includes:

## Version Control Rules
- All dependencies must be pinned to exact versions in package.json
- Claude defaults to its training data; pinned versions prevent silent drift

## Security Requirements
- API keys never touch the client environment
- All 3rd party credentials stay server-side
- Bearer security scanner runs on every PR
- Critical/High findings block merge

## Build Gates
- No sprint begins without locked AI Spec
- All spec questions must be resolved pre-build
- Stakeholder signoff required at Phase 3 boundary

These rules ship in every project's CLAUDE.md from day one, enforcing the framework's methodology through Claude Code's own configuration system.

The Elicitation Method — Pre-fill Synthesis, Elicit Discovery

CRISP's questioning technique is specifically designed for AI projects: "Elicit, don't interrogate. Every question follows the same pattern: pre-fill from what you already know, present a hypothesis, let the client correct. People react better than they originate. Corrections surface real intent faster than open questions."

For documents that compile prior decisions, CRISP templates are pre-filled and confirmed. For documents that surface new information (stakeholder impacts, hidden risks, unstated goals), the framework provides active elicitation moves before confirmation.

Try It Now — Your Next Client Project

Instead of jumping straight to claude code "build a dashboard for sales data", try:

1. Run through CRISP Phase 1 (Problem Definition) using the template
2. Fill the Technical Stack Table with your actual versions
3. Generate the AI Implementation Spec with all questions resolved
4. Drop the resulting CLAUDE.md into your project
5. Only then: claude code "implement according to docs/ai_spec.md"

The framework's GitHub repository includes complete examples showing how each phase flows into the next, with explicit data lineage showing where each requirement originated.