ContextQMD
Back to Sanity

LLM adaptation prompt

examples/functions/stale-content-digest/PROMPT.md

5.31.13.1 KB
Original Source

LLM adaptation prompt

Copy everything below the line into your LLM (Claude Code, Cursor, Codex, ChatGPT, Claude.ai with the Sanity MCP, etc.). Fill in TARGET DOCUMENT TYPE and TARGET FIELD first.

For coding agents running in your project root, AGENTS.md gives them this same context automatically — you can also just say "adapt this function to my [type] content" and the agent will follow the steps below.

TARGET DOCUMENT TYPE: <fill in, e.g. "documentation"> TARGET FIELDS: <fill in one or more comma-separated fields to review for staleness, e.g. "body" or "title, lead, body">

I want to adapt the stale-content-digest Sanity Function in this project to work with the document type and field(s) above. The function currently targets the movie document type and the single overview Portable Text field — rewire it for my schema.

Steps:

  1. Verify the target. Confirm the document type and every field I named above actually exists in my Sanity schema. Use list_workspace_schemas and get_schema from the Sanity MCP if available, otherwise read the schema files (typically under schemaTypes/ or schemas/). For each field, note its type (string, text, Portable Text array, array of strings, etc.) — you'll need it for the GROQ projection. If the type or any field doesn't exist, stop and ask me to correct it.

  2. Edit stale-content-digest/index.ts in three places — the same three points called out in the README's "Customization" section:

    a. The GROQ query (around the STALE_QUERY constant). Change _type == "movie" to my type. Project the title-equivalent field and each target field. Choose the projection per field based on its type:

    • Plain string or text: myField (or rename)
    • Portable Text array: "myField": pt::text(myField)
    • Array of strings: "myField": array::join(myField, " ")
    • Anything else (object, reference, etc.): stop and ask me how I want it flattened to text

    If I gave you multiple fields, project each one separately under its own key (don't pre-concatenate in GROQ). The agent prompt and the TypeScript interface will reflect them all.

    b. The TypeScript interface at the top of the file (currently Movie). Rename it and update the field types to match what the query returns.

    c. The agent instruction prompt (the instruction field passed to the agent action). Replace the snarky-reviewer persona with one tailored to my content type. Be explicit about what staleness looks like for my domain — e.g., outdated API references for docs, expired promotions for marketing content, broken links for resource pages. Keep the JSON output schema unchanged.

  3. Update the user-facing strings in index.ts — log messages and the Slack message header — so they say "stale [my content type]" instead of "stale movies".

  4. Update the Studio deep-link in formatSlackMessage if the document type slug differs from movie.

After editing, show me a diff and tell me what to set DAYS_SINCE to for a useful first run on my content.