Two files. Five minutes. Declare your stack, paste a transcript, and get governed build specs — one per feature, validated before you build. No infrastructure, no code, no credit card.
The Rikki Skill is two markdown files — the instructions that make Claude behave as Rikki, and the governance references it draws from.
Both files are plain text — you can open them in any editor to review before uploading.Go to claude.ai → Projects → New project. Name it something like "Rikki — Rikki Skill" so you can find it easily.
Inside your project, click "Set project instructions". Open OG-SKILL-INSTRUCTIONS.md, copy the entire contents, and paste it into the instructions field. Save.
This is what makes Claude behave as Rikki — do not skip this step.In the Project Knowledge section, click "Add content" and upload GOVERNANCE-REFERENCES.md. This gives Rikki the framework citations it applies to your artifacts.
Start a new conversation inside the project. First, tell Rikki your tech stack — this becomes the baseline for every spec it generates. Then paste a transcript and Rikki will interpret the intent, generate governed specs (one per feature), score confidence, and surface gaps.
The Claude Project version is optimized for Claude. A universal prompt version that works across ChatGPT, Gemini, and other LLMs is coming soon — join the waitlist on the main site to be notified.
You do. Rikki generates the artifacts from your conversations — they belong to your organization. The skill files are yours to use, modify, and share.
Your conversations are governed by Anthropic's privacy policy for Claude.ai. Rikki is designed to treat all conversation content as confidential and does not carry context between sessions.
OWASP Top 10, OWASP LLM Top 10, NIST AI RMF, ISO 42001, ISO 27001, NIST CSF, WCAG 2.2, EU AI Act, STRIDE, and MITRE ATLAS — but only what's genuinely relevant to the feature scope. A static page won't get STRIDE analysis. A payment flow will.
Yes. Paste any spec (one Rikki generated, or your own) and say "validate this before we build." Rikki will review it as an adversarial reviewer — surfacing blockers, risks, ambiguous requirements, missing acceptance criteria, and governance gaps. It outputs a Validation Report with a clear verdict: clear to build, build with caution, or do not build yet.
At the start of a conversation, tell Rikki your stack: "Our stack is React, Node.js, PostgreSQL, AWS." From that point, Rikki treats your stack as confirmed constraints (🟢) across every spec it generates. It will flag any requirement that conflicts with your stack, and tailor governance recommendations to your specific technologies.
The free Rikki Skill works in Claude Projects with no infrastructure. Rikki Platform (coming soon) adds persistent organizational memory, enterprise data connectors (Jira, GitHub, Slack, Linear), cross-session change tracking, and spec-to-code alignment. Join the waitlist to be first to know.