For agents

The guidance pages reproduce shipped files from rcf-lite@0.36.1, installed in the site's node_modules. The MCP surface and Cloudflare-first guide are source-audited editorial maps. The docs build reads the pack at build time. A drift test in CI rejects changes to the shipped bytes that differ from the site's checked-in copy. The guidance text on the Harness block and Persona programme pages is directly projected from the same files served through MCP, so their public URLs provide a fallback without MCP. The Elicitation playbook, Build-cycle playbook and Document model pages add editorial corrections that are not present in the MCP-served pack content.

• Harness block In a fresh repository, rcf init writes the managed block into both CLAUDE.md and AGENTS.md. When either file already exists, it refreshes CLAUDE.md preferentially, otherwise AGENTS.md. The operating rules your agent follows, verbatim. →
• Elicitation playbook The playbook the agent follows before writing a document. The questions it asks and how your answers become the chain. →
• Build-cycle playbook The five-stage loop the agent runs for every build item: define, build, review, test, finalise. Each stage commits. →
• Document model The types the chain uses, their fields and relationships. This page is a document-model overview. For authoritative fields, enums and ID rules, use the per-schema references shipped with @stravica-ai/rcf-schemas. →
• Persona programme The tail-interview template, dispatched in a dedicated turn as the mandatory close of every persona run. →
• MCP surface The tools and prompts the rcf MCP server exposes to a connected agent, one entry each. →
• Cloudflare-first projects When to use the Cloudflare blueprints, how they compose across deployment, persistence, scheduling and edge security, and when to connect a Hetzner backend through a Cloudflare tunnel. →

Start with the harness block to read the operating rules your agent follows.