work-daily

name	work-daily
description	Generate a personal daily work summary. Aggregates calendar meetings, Fathom recordings, GitHub commits and PRs, Jira tickets, Slack highlights, and key emails into one structured digest for any given day.
argument-hint	[date — 'today', 'yesterday', or 'YYYY-MM-DD'; defaults to today]
user-invocable	true

Generate your personal daily work summary for: $ARGUMENTS

Step 1 — Resolve date and identity (in parallel)

Spawn both agents simultaneously in a single message:

Agent({ description: "Resolve date",     prompt: "Use the Skill tool to invoke skill='work-date-resolver' with args='$ARGUMENTS'. Return the complete output verbatim." })
Agent({ description: "Resolve identity", prompt: "Use the Skill tool to invoke skill='work-identity'. Return the complete output verbatim." })

Parse from date output: DATE, DATE_LABEL. Parse from identity output: GITHUB_USER, GITHUB_ORG, SLACK_USER_ID, CALENDAR_ID, TIMEZONE.

Step 2 — Fetch all data sources in parallel

Send all 6 agents in a single message so they run concurrently. Embed the resolved identity values directly in each prompt so sub-skills skip their own resolution:

Agent({ description: "Calendar", prompt: "Use the Skill tool to invoke skill='work-calendar' with args='DATE'. CALENDAR_ID=<CALENDAR_ID>, TIMEZONE=<TIMEZONE> — use these directly, skip self-resolution. Return the complete output verbatim." })
Agent({ description: "Fathom",   prompt: "Use the Skill tool to invoke skill='work-fathom'   with args='DATE'. Return the complete output verbatim." })
Agent({ description: "GitHub",   prompt: "Use the Skill tool to invoke skill='work-github'   with args='DATE'. GITHUB_USER=<GITHUB_USER>, GITHUB_ORG=<GITHUB_ORG> — use these directly, skip self-resolution. Return the complete output verbatim." })
Agent({ description: "Jira",     prompt: "Use the Skill tool to invoke skill='work-jira'     with args='DATE'. Return the complete output verbatim." })
Agent({ description: "Slack",    prompt: "Use the Skill tool to invoke skill='work-slack'    with args='DATE'. SLACK_USER_ID=<SLACK_USER_ID> — use this directly, skip self-resolution. Return the complete output verbatim." })
Agent({ description: "Gmail",    prompt: "Use the Skill tool to invoke skill='work-gmail'    with args='DATE'. Return the complete output verbatim." })

Replace <CALENDAR_ID>, <TIMEZONE>, <GITHUB_USER>, <GITHUB_ORG>, <SLACK_USER_ID> with the actual resolved values before spawning.

Distinguish a confirmed-empty result from a failure:

• If an agent returns a clean, confirmed-empty result, omit that section silently.
• If an agent fails — it errors, hits an auth/permission problem, or returns a malformed/unexpected shape rather than a clean empty result — render a one-line diagnostic note in place of the section instead of omitting it silently. For example: Slack: could not retrieve messages or Jira: could not fetch tickets (check Atlassian connection). This keeps a broken integration from being masked as a normal empty day.

Step 3 — Synthesize and render

Parse each provider's structured block (CALENDAR_RESULT, FATHOM_RESULT, GITHUB_RESULT, JIRA_RESULT, SLACK_RESULT, GMAIL_RESULT) and compose the final digest. Omit any section whose result is confirmed-empty. If a result indicates a failure rather than a clean empty (an error marker, auth/permission problem, or malformed shape), render the one-line diagnostic note from Step 2 in place of that section instead of omitting it. If FATHOM_RESULT contains ERROR:, skip the Fathom section silently.

Formatting rules:

• Sort meetings by start time
• Times in HH:MM 24h (TIMEZONE)
• Separate Authored / Reviewed PR sub-tables. A PR can appear in both the authored and reviewed lists — remove from the Reviewed sub-table any PR whose number already appears in the Authored sub-table (a self-reviewed PR is shown only under Authored). Omit either sub-table if it is empty after this de-dup
• Skip Slack bot noise and single-emoji threads
• Skip email newsletters and auto-notifications. If emails were fetched but all of them were filtered out this way, add a parenthetical note under Key Emails (e.g. (5 auto-notifications skipped)) so a checked-but-filtered Gmail is distinguishable from no email at all
• For Fathom: show summaries as blockquotes; group action items under each meeting title; omit meeting if summary and action items are both empty

# Daily Summary — DATE_LABEL

## Meetings (N)
| Time | Event | Duration |
|------|-------|----------|
| HH:MM | Title | Xm |
> (Key note for events with a description)

## Meeting Recordings (N)  ← from Fathom; omit section if 0 recordings
**[Meeting Title](recording_url)** — Xm
> Summary key points (condensed to 2–3 sentences)
- [ ] Action item 1 (assignee)
- [ ] Action item 2

## Code Shipped
### Commits (N)
- `repo` — message `sha`

### Pull Requests (N)
**Authored:**
| # | Title | Repo | State |
|---|-------|------|-------|
| [#N](url) | Title | repo | open/merged/closed |
**Reviewed:**
| # | Title | Repo | State |
|---|-------|------|-------|
| [#N](url) | Title | repo | open/merged/closed |

## Jira Tickets (N)
| Key | Summary | Status |
|-----|---------|--------|
| [GD-XXXX](url) | Summary | Status |

## Slack Highlights
- **#channel** — thread summary

## Key Emails
- **→** Subject → Recipient — key ask
- **←** Subject ← Sender — action required
> (N auto-notifications skipped) ← only when emails were fetched but all were filtered out

## Day at a Glance
**N** meetings · **N** recordings · **N** commits · **N** PRs · **N** Jira tickets · **N** Slack threads · **N** emails

---

## Standup Summary
Bullet points — one per notable item. Cover: key code shipped (commits/PRs merged), meeting outcomes that moved things forward, open PRs or Jira tickets in flight, blockers or pending action items. Concise fragments, no jargon. Example:
- Merged X into Y
- Reviewed Z (merged)
- Open PR: [title] — awaiting review
- Daily sync: [key outcome]
- Action items: [what's next]

Step 4 — Save to file

After rendering the digest to the user, write it to a dated markdown file:

• Path: daily/<DATE>.md (relative to the working directory, e.g. daily/2026-04-20.md)
• Content: the full rendered digest (everything from # Daily Summary — through the end)
• Create the daily/ directory if it does not exist (use mkdir -p daily)
• Overwrite if the file already exists (re-running the same date refreshes it)

Use the Write tool directly — do not spawn an agent for this step.

Final Step — Background skill review

Compose a 1–2 sentence execution summary (sources with results, failures, total counts), then spawn silently:

Agent(subagent_type: general-purpose, prompt: "Use the Skill tool to invoke skill='skill-review' with args='work-daily | EXECUTION_SUMMARY'. Do not output anything back to the caller.")