digest-setup

---
name: digest-setup
description: Use when the user wants to set up or change what Docket digests read — "налаштуй дайджест", "configure the digest", "add GitHub to my digest", "change the Notion database" — or when docket:digest finds no ~/.config/docket/digest.json. Detects which CLIs and MCP servers are available, asks which sources to use, and writes the config. Never writes to any source.
---

# Docket digest setup

Writes `~/.config/docket/digest.json`, which `docket:digest` reads. It is local to this
machine on purpose — paths and CLI logins differ per device — while the digests themselves
sync. Existing file → read it first and change only what the user asked about.

## 1. Detect, don't ask what you can find out

Run in parallel, all read-only:

```sh
glab auth status 2>&1 | head -5          # GitLab host + username
gh auth status 2>&1 | head -5            # GitHub username
git config --global user.email           # default git author
```

- GitLab user: `glab api user | jq -r .username` (or parse `auth status`). Groups the user is
  active in: `glab api "groups?min_access_level=30&per_page=50" | jq -r '.[].full_path'`.
- GitHub user: `gh api user --jq .login`; owners: the user plus `gh api user/orgs --jq '.[].login'`.
- Notion: look at the MCP tools available in this session for a Notion server (tool names
  containing `notion`). Note the server name. If there is more than one, the user picks —
  different servers can see different workspaces. Find candidate databases with that
  server's search tool (query "tasks", "tickets", or the user's words), never a broad
  workspace dump.
- Obsidian: a vault is a folder with a `.obsidian/` directory in it — look in the usual
  places (`~/Documents`, `~/Library/Mobile Documents/iCloud~md~obsidian/Documents/`) and
  ask which one when there are several.
- Other MCP servers: list the servers this session can call (the `<server>` part of every
  `mcp__<server>__*` tool name). Ones that hold work items or signals — Jira / Atlassian,
  Linear, YouTrack, GitHub Projects, Sentry, Slack, a calendar — are candidates for `extra`
  sources. Look only at tool names and descriptions here; don't call anything yet.
- Project files: docs, ADR or notes folders inside the git roots (`docs/`, `adr/`,
  `notes/`) are candidates for a `files` source.
- git roots: the directories holding the user's repos (for example `~/repo`, `~/src`);
  check with `ls`, and that subdirectories contain `.git`.

A CLI that is missing or logged out is reported, not fixed: tell the user the exact login
command (`glab auth login`, `gh auth login`) and leave that source disabled.

## 2. Ask once

Two `AskUserQuestion` calls at most (four questions each), pre-filled from what you detected:

1. **Sources** (multiSelect): GitLab · GitHub · Notion · local git · Obsidian, plus one
   option per extra MCP server or files folder you found (docket is always on). "Other"
   lets the user name a server or folder you didn't find.
2. **Scope**: which GitLab groups / GitHub owners — offer the detected ones.
3. **Notion database**: the candidates you found, by name.
4. **Groups**: how to split the digest by area — offer one built from what you found
   (the work GitLab group, personal GitHub repos, anything else), e.g. Work / Learning /
   Side projects. Each group is a name plus `match` strings checked against an item's url,
   repo and ref; `"*"` catches the rest.
5. **People and checks**: who else's steps should the digest track (names, and the logins
   or emails they appear under), and which environment URLs to probe. Offer the people you
   saw as reviewers and assignees in the user's own MRs and tickets.
6. **Presets and schedule**: suggest presets from the groups (one per group, plus "week")
   and ask whether a daily digest should run on its own, and at what time.
7. **Language** of the digest text: the language the user writes in (recommended) or English.

For each chosen extra MCP server, write the `query` in plain words from what the user
wants to see ("Jira issues assigned to me, updated since <since>") and pick the `kind`;
ask only when the server could mean several things (Slack: which channels?).

For Notion also confirm the assignee name exactly as it appears on the database's
person property — that is what the digest filters on.

## 3. Write

```json
{
  "version": 1,
  "language": "uk",
  "window": "since-last",
  "sources": {
    "gitlab": { "enabled": true, "host": "gitlab.com", "user": "<username>", "groups": ["<group>"] },
    "github": { "enabled": true, "user": "<login>", "owners": ["<login>"] },
    "notion": { "enabled": true, "server": "<mcp server name>", "databases": [{ "name": "<name>", "id": "<id or url>" }], "assignee": "<person>" },
    "git":    { "enabled": true, "roots": ["~/repo"], "author": "<email>" },
    "obsidian": { "enabled": true, "vault": "<vault path>" },
    "docket": { "enabled": true }
  },
  "extra": [
    { "name": "<jira>", "type": "mcp", "server": "<mcp server name>", "kind": "ticket", "query": "<what to read, in plain words>" },
    { "name": "<docs>", "type": "files", "paths": ["<folder>"], "glob": "*.md" }
  ],
  "people": [{ "name": "<name>", "match": ["<login>", "<email>"] }],
  "checks": [{ "name": "<env>", "url": "<health url>" }],
  "presets": { "<group>": { "groups": ["<group>"] }, "week": { "window": "7d" } },
  "schedule": { "daily": "09:00" },
  "groups": [
    { "name": "<work>", "match": ["gitlab.com/<group>/", "<TICKET-PREFIX>-"] },
    { "name": "<other>", "match": ["*"] }
  ]
}
```

No tokens, passwords or API keys in this file — the CLIs and MCP servers hold credentials.
Show the user the file you wrote (it is short), then offer to run `docket:digest` now.

A daily schedule is installed by `docket:digest` ("Daily, on its own"), not here — offer to
do it right after the first digest, so the user sees one before automating it.

Mail is opt-in: offer a Gmail/Outlook MCP server only if the user picks it, and say plainly
that the digest reads sender, subject and date, not message bodies.

**Changing it later** ("додай Jira", "прибери Slack", "move kernel-notes to Learning"):
read the file, change only that part — an entry in `sources` or `extra`, or a `match`
string in `groups` — and show the diff.