> ## Documentation Index > Fetch the complete documentation index at: https://docs.getmillwork.dev/llms.txt > Use this file to discover all available pages before exploring further. # Use a coding assistant > Give any coding assistant the right Millwork context without sharing secrets or unintended authority. Millwork documentation is written for people and coding assistants. You can use Codex, Claude, Cursor, Gemini, or another assistant. The workflow is the same. ## Give the assistant the page Open the **Copy page** menu at the top of a documentation page. Copy the page as Markdown, or choose **View as Markdown** and share that URL. After the documentation site is published, an assistant can also discover the current public documentation through: * `/llms.txt` for the page index; * `/llms-full.txt` for the complete public documentation; and * `/skill.md` for Millwork capabilities and safety limits. ## Start with one request Tell the assistant the outcome you want. It should collect the safe technical context itself instead of asking you to complete a form. ```text theme={null} Help me complete this Millwork documentation task. Documentation: Outcome I want: Rules: - Follow the documentation exactly. Keep endpoint, header, and field names unchanged. - Inspect the current non-secret project files and terminal output to collect the facts you need. Do not ask me to fill in a checklist. - Do not open or print .env files, secret stores, browser storage, cookies, API keys, provider credentials, secure setup links, existing prompts, existing model output, or customer data. For an approved documented run using public information or test data created for the run, keep the new result local and report only status, provenance, and safe counts. - Use placeholders for every secret. - Before each command, tell me what it reads or changes. - Do not send email or make any API, dashboard, or provider request that creates, changes, cancels, rotates, revokes, or deletes state without asking me first. This includes free test runs and connection tests. Read-only checks and local checks are allowed. Never start paid work without separate approval. - Run one bounded check at a time. Do not retry or fall back unless the page says to. - Compare each result with the page's Expected result. - If a step fails, stop and return the HTTP status, problem type, title, instance, and one short redacted error excerpt. Do not guess. - If one required fact is still missing, ask one short question. Otherwise continue. Return: 1. prerequisites; 2. the next command or action; 3. the expected result; 4. what to do if it fails; and 5. any step that needs my approval. ``` ## Share only what is needed Share diagnostic fields only with an assistant your organization permits to access this project. | Usually safe diagnostic context | Do not share | | ------------------------------------ | --------------------------------------------- | | Documentation URL or copied Markdown | Passwords, tokens, API keys, or cookies | | Endpoint and HTTP method | Authorization headers or provider credentials | | HTTP status and problem fields | Secure setup or invitation links | | Run or connection ID | Prompts, model output, or customer data | | Short redacted error excerpt | Full logs, request bodies, or response bodies | ## Keep authority with you A coding assistant may read documentation, inspect non-secret project files, collect safe context, draft commands, run read-only or local checks, and explain results. It should do those steps without turning them into a form for you. It must ask before any remote write or other material action, including submitting or cancelling a run, testing or changing a provider connection, creating or revoking a key, sending a message, starting paid work, or deleting data. Use a bounded example with an expected result and recovery step. Let an assistant draft a safe report for you to review and send.