/crisp-ai
Review chat, agent, and other AI interfaces.
(What it does)
A design standard for AI interfaces. /crisp-ai reviews chat interfaces, streaming responses, AI agents, generative UI (interface the AI builds on the fly), and inline assistants against six AI-specific dimensions. These cover problems that did not exist before 2023, such as showing uncertainty without meaningless percentages and keeping streaming text readable.
Each dimension links back to the core CRISP framework, is scored out of ten, and comes with named problems and exact fixes. You also get component-level recommendations, compared with products like Claude.ai, Perplexity, and Cursor.
(When to use it)
Use it when AI is a main part of the experience, not an add-on:
- Building or reviewing a chat interface, an agent, or generative UI
- Adding an AI feature to an existing product and deciding how it should behave when it is wrong
- Streaming output feels disorienting and you need to explain why
- Users either trust the AI too much or ignore it, and nothing in the interface helps them judge
(What you get)
- Scorecard across 6 AI dimensions
- Streaming UX recommendations
- Guidance on agent feedback patterns
- Error and uncertainty handling
(Example output)
Illustrative example only. This is not a real audit.
(The six dimensions)
- Uncertainty communication: does the user know when to trust the output and when to check it?
- Streaming legibility: can the user read the output as it arrives?
- Failure grace: when the AI fails, is the way to recover clear, without blaming the user?
- Human override clarity: can the user easily correct, edit, or undo AI output?
- Context transparency: does the user know what information the AI is using?
- Progressive AI power disclosure: are capabilities introduced gradually, without overwhelming the user?
(Common questions)
How does this relate to /crisp-audit?
They are complementary scorecards. /crisp-audit scores the five core dimensions and /crisp-ai scores the six AI-specific ones. An AI-native product needs both. If .crisp.md marks the product as AI-Native, this review runs automatically as part of the full audit.
What surfaces does it cover?
Chat interfaces, inline assistants inside editors, autonomous agents, generative UI, AI-powered search, and conventional products with AI built in. It asks which type you are working on first, because each one fails in different ways.
Why not just show confidence percentages?
Because users do not know how to interpret them. The uncertainty dimension recommends qualitative patterns instead: source citations, prompts to verify, visually de-emphasising generated content, and showing the AI's knowledge cut-off when it matters.
Does it assume the AI will fail?
Yes, on purpose. AI fails in particular ways: running out of context, hitting content policy, reaching the limits of its knowledge, or being confidently wrong. Each needs its own recovery path. The standard is to design honestly for failure, not to hide it.
(Install)
Installs all fourteen CRISP skills and detects which AI coding tool you use. Run /crisp-teach once per project, before the other skills.
(Related skills)