CRISP plugin
The CRISP method as skills: five phases from a vague client brief to a spec Claude Code can start a sprint from.
- 53 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The skills set a work order and write documents in `docs/`, calling no external services themselves.
Install
- Works in
Manual install
Copy the `.claude/skills` folder into the Claude Code project, run `/crisp-orchestrator` and complete phases in order, checking each exit list.
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skills set a work order and write documents in `docs/`, calling no external services themselves.
This is third-party code. Review the repository files before installing.
What it does
CRISP is a method for AI implementors and analysts. A set of skills, `/crisp-orchestrator`, `crisp-start`, `crisp-execute` and `crisp-report`, carries a project through five phases. Clarify defines the problem and constraints and weighs buy versus build. Results records stakeholders and baseline metrics. Investigate maps processes and user journeys. Spec assembles a stack with pinned versions, a backlog and an AI spec for each sprint. Prove checks the outcome against the baseline. Project state lives in one JSON file in `docs/` that every phase reads and updates, and the final `CLAUDE.md` is compiled from all decisions. Security rules such as keeping API keys server-side enter the project from day one.
Who it is for. AI agency owners, consultants and product managers who turn a client request into a spec for Claude Code.
Good fit when
- A client brings a vague idea and the problem and success metrics must be defined before code.
- A team chooses between buying and building with a market assessment.
- A process map, stakeholder roles and a backlog are assembled for a project.
- An AI spec without open questions is written before a sprint.
Not a fit when
- The task is small and writing code right away is enough.
- You need a code framework rather than a method.
- The project has no client or business metrics.
Example request
Run /crisp-orchestrator for a new project: the client wants an AI support assistant, help define the problem and metrics.Limitations
The method is demanding: phases run in order without skipping and an exit checklist is verified before each transition. Bearer security scanning on every pull request is part of the project rules and needs separate setup. Some steps, such as market and competitor research, use external sources and their findings should be verified. Results are written into your project `docs/` folder and the blank templates must not be overwritten.
How to disable. Remove the `.claude/skills` folder from the project and the `docs/` files created by the phases. Edit the master `CLAUDE.md` by hand.
FAQ
Where do I start with CRISP?
Put the `.claude/skills` folder into the Claude Code project and call `/crisp-orchestrator`. It detects where you are and continues from the right phase.
What do the letters of CRISP stand for?
Clarify, Results, Investigate, Spec, Prove.
What comes out of the Spec phase?
A stack with pinned versions, requirements, a backlog, an AI spec and `CLAUDE.md`. After that, just say "Start Sprint 1".
Why pin versions?
Without pinned versions Claude relies on its training data and drifts silently. Versions go into the stack table and into `CLAUDE.md`.
Is there an example project?
Yes, `examples/longevity-platform/` holds an example with a `CLAUDE.md` and a buy versus build matrix.
Discussion
No questions or reviews yet.
Related
PluginsA documentation-first method for Claude Code: the /interview skill, document and plan templates and handoff rules.
The claude-crusts CLI splits a Claude Code context into six categories, finds waste and writes a /compact command for the session.
A CLAUDE.md and AGENTS.md file with seven principles: the agent frames the problem and scope before it writes code
Claude Code course repository for product managers: 55 skills, slash commands, subagents and demo apps with module assignments
A Claude Code plugin that catches your corrections in chat and, after review, writes them into CLAUDE.md and AGENTS.md
A skill that keeps project memory in plain Markdown files: session logs, a wiki, rules, a to-do list and a note on where to resume
README in short
The README explains five phases, security rules and the working order. The human entry point is `START_HERE.md`, and a setup video exists.