InFlow
Skill pack

Docs

AI skill pack

A shareable pack of prompts and a SKILL.md that teaches an agent to work with your vault file. You attach inflow-db.json. You run the model. InFlow does not.

What it is

Files you can download from the skill pack page: SKILL.md (agent instructions), schema.md (vault shape), prompts.md (copy-paste), and a zip of all three.

It is not an InFlow chatbot, not an API, and not a model we host. Schema fields for a future partner layer stay empty in this app release.

How to use it

  1. 1Get a copy of inflow-db.json — from Google Drive (InFlow/inflow-db.json) or from a backup ZIP. That file is yours; treat it as sensitive.
  2. 2Install SKILL.md in Claude / Cursor / Codex / Grok skills, or paste a prompt from the pack into ChatGPT or any chat that accepts file attachments.
  3. 3Attach the JSON. Ask for a weekly review, inbox triage, recall, or a schema-safe patch.
  4. 4Read the answer. If you want edits in the app, apply them yourself (or merge a proposed JSON) — then let InFlow sync as usual.

What the skill is allowed to do

  • Read collections: folders, views, entries, tags, types
  • Quote titles and ids that exist in the file
  • Suggest workspace moves and existing tags
  • Propose new entry objects that match the schema
  • Emit a patch or diff, not a silent rewrite of the whole vault

What it must not do

  • Invent entries, workspaces, or “memories” that are not in the file
  • Hard-delete rows or Drive attachment ids
  • Write vectors or embeddings into inflow-db.json
  • Bump schemaVersion or drop unknown keys
  • Claim that InFlow ran the model

Attachments

The JSON holds metadata (mediaFileId, mime, duration). Photo, voice, and video bytes live in InFlow/attachments/ or on the device. A model that only has the JSON cannot watch the video. Export a backup ZIP if you also want those files in the same folder — still your copy, still not uploaded to us.