hourly-digest git:20260611.00e1073A

Immutable. This exact content never changes and is served at /api/v1/blob/bb78a71fa571116a.

---
name: hourly-digest
description: |
  This skill should be used when the user asks to "check hourly updates",
  "hourly digest", "slack hourly summary", "gmail daily summary",
  "what happened in the last hour", "hourly monitoring", or needs a periodic
  digest from Slack or Gmail. Runs as two independent loops on different
  cadences: Slack hourly, Gmail daily. Also triggers for automated invocation
  via /loop 1h /hourly-digest:hourly-digest slack and
  /loop 1d /hourly-digest:hourly-digest gmail.
  Distinct from per-channel Slack digest or single-source summaries.
  For GitHub/Linear activity sync, use /loop 1h /github-activity separately.
argument-hint: "[slack|gmail]"
---

# Hourly Digest

Two independent digest loops on different cadences: Slack messages from the past hour, Gmail from the past day. Respond in the user's language.

The cadences differ by source constraint: Gmail's `newer_than:` search operator only supports day/month/year units (no hour granularity), so the Gmail loop runs daily while the Slack loop runs hourly.

## Argument Parsing

Parse `ARGUMENTS` for an optional subcommand:

| Input | Behavior |
|-------|----------|
| (empty) | Run both sections: Slack (past hour), Gmail (past day) |
| `slack` | Slack hourly digest only |
| `gmail` | Gmail daily digest only |

## Execution

Run each requested section independently. On failure, report `[ERROR] Source: {message}` and continue to the next section.

Load the MCP tools for the requested sections upfront in a single ToolSearch call — only the tools the requested sections need, so a missing connector can only fail its own section:

```
Slack section:  ToolSearch("select:mcp__claude_ai_Slack__slack_search_public_and_private")
Gmail section:  ToolSearch("select:mcp__claude_ai_Gmail__gmail_search_messages,mcp__claude_ai_Gmail__gmail_read_message")
Both sections:  one ToolSearch call selecting all three tools
```

### 1. Slack Hourly Digest (past hour)

1. Compute today's date and the Unix timestamp of one hour ago via Bash: `date '+%Y-%m-%d'` and `date -v-1H +%s` (Slack query `after:` excludes the given date; use `on:` to include it)
2. Use `slack_search_public_and_private` with query `on:{date}` plus parameters `after={unix_ts_one_hour_ago}`, `sort="timestamp"`, `sort_dir="desc"` — the `after` parameter windows results server-side, so result-count truncation cannot drop in-window messages; paginate with `cursor` while a full page (20 results) comes back. During the first hour of the day (00:00-00:59), also run a second query with `on:{yesterday}` and the same `after` timestamp so the window spanning midnight is covered
3. Group results by topic — cluster messages about the same subject together, even when they span multiple channels
4. Per topic: `**Topic** (N): key points — [#channel-name](https://{workspace}.slack.com/archives/{channel_id})`
5. For notable messages with threads, link to the thread: `[thread](https://{workspace}.slack.com/archives/{channel_id}/p{timestamp_no_dot})`
6. If no messages found, state that there are none

Construct Slack links from search result metadata: channel ID for channel links, channel ID + message timestamp (remove the dot) for thread permalinks.

### 2. Gmail Daily Digest (past day)

1. Use `gmail_search_messages` with query `newer_than:1d`
2. If search results already include sender and subject, use them directly. Only call `gmail_read_message` for items missing these details (avoid N+1 reads)
3. Group by topic — cluster related subjects and threads together
4. Per topic: `**Topic** (N): key points — ["Subject"](https://mail.google.com/mail/u/0/#all/{messageId}) from sender`, listing a link for every message in the topic (do not collapse to a single link when N > 1)
5. If no emails found, state that there are none

Construct Gmail links from the messageId returned by search results. Use the `#all/` (All Mail) view in link URLs — `#inbox/` breaks for archived, sent, or labeled-out messages that the search still returns.

## Output Format

Each section gets its own header carrying its own time window:

- Slack: `## Slack Hourly Digest (HH:MM-HH:MM KST)`
- Gmail: `## Gmail Daily Digest (past 24h, as of YYYY-MM-DD HH:MM KST)` — `newer_than:1d` is a rolling 24-hour window, not a calendar day

Example structure:

```
## Slack Hourly Digest (14:00-15:00 KST)
- **Auth service hotfix** (3): deployment discussion and rollout decision — [#channel-a](https://workspace.slack.com/archives/C0123), [hotfix thread](https://workspace.slack.com/archives/C0123/p1234567890)
- **Standup** (1): reminder — [#channel-b](https://workspace.slack.com/archives/C0456)

## Gmail Daily Digest (past 24h, as of 2026-06-12 15:00 KST)
- **Billing renewal** (1): ["Subject line"](https://mail.google.com/mail/u/0/#all/msg123) from sender@example.com
- **Project kickoff** (2): ["Thread topic"](https://mail.google.com/mail/u/0/#all/msg456) from another@example.com, ["Re: Thread topic"](https://mail.google.com/mail/u/0/#all/msg789) from third@example.com
```

When a section has no results, use a single line indicating nothing new.
When running a single section via subcommand, omit the other section entirely.

## Cron Usage

```
/loop 1h /hourly-digest:hourly-digest slack
/loop 1d /hourly-digest:hourly-digest gmail
```

Plugin skills are namespaced as `{plugin-name}:{skill-name}`; the bare `/hourly-digest` form only resolves for a user-level skill installation.