agentleFS
serhii12serhii12/vault-os/.claude/skills/daily-recap/SKILL.md

daily-recap

Skill2 starsChanged 6 months agoMITView on GitHub

Reconstruct a chronological timeline of a working day from external activity — Slack, GitHub, calendar, meeting notes (Granola), and Claude Code sessions. Produces a markdown file with stand-up block, timeline, and stats. Use when the user says "recap", "/daily-recap", "what did I do today", "yesterday recap", or asks for a retrospective of a day's work.

3 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 5 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use the argument if provided, otherwise today.

    line 13
  2. Distinctive:

    Never guess the day-of-week — always compute it.

    line 20
  3. Distinctive:

    Use those timestamps as the source of truth.

    line 61
  4. Distinctive:

    No meeting-by-meeting detail or Slack chatter — anything that wouldn't belong in a standup gets dropped.

    line 141
  5. Very specific:

    Write to _agent/outputs/recaps/<YYYY-MM-DD> - <DayOfWeek> - Recap.md (e.g., 2026-05-06 - Wednesday - Recap.md).

    line 145
  6. Distinctive:

    If the file exists, read it first, merge in new entries, and preserve any manual edits.

    line 147

“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Daily Recap — Retrospective Timeline

Reconstructs what happened during a working day from external sources. Pairs with close-day (which processes the user's own daily-note writing).

Date

Use the argument if provided, otherwise today.

TARGET_DATE="${ARG:-$(date +%F)}"
DAY_NAME=$(date -d "$TARGET_DATE" +%A)   # macOS: date -j -f "%Y-%m-%d" "$TARGET_DATE" +%A

Never guess the day-of-week — always compute it.

Conventions

  • Vault root: operate from CWD. The skill auto-detects the Claude project directory matching the current path.
  • Output: _agent/outputs/recaps/<YYYY-MM-DD> - <DayOfWeek> - Recap.md. The directory _agent/outputs/ exists; create recaps/ on first run.
  • Timezone: defaults to the user's local TZ from the shell. Override with TZ env var. The "working day" boundary defaults to 03:00 local time on the target date through 03:00 local time the next day, capturing late-night sessions.

Prerequisites

Required:

  • Google Calendar MCP — for events.
  • gh CLI authenticated — for GitHub events / PR enrichment. Skip GitHub steps if gh auth status fails.

Optional (skill degrades gracefully if absent):

  • Slack MCP — at session start, scan available tools for mcp__*[Ss]lack* (e.g., search_messages, read_channel). If none, skip Slack steps and note in output.
  • Granola MCP (mcp__*Granola*) — for meeting notes / actual durations.
  • Claude Code session logs at ~/.claude/projects/<slugified-vault-path>/*.jsonl — used for "what was worked on" attribution.

Step 1 — Gather data (parallel)

Run these concurrently:

  1. Slack messages from me — search from:me on:<date>, sorted ascending, paginate. Also search from:me on:<next-date> and keep only messages before 03:00 local time.
  2. Slack messages to me — search to:me on:<date> (same logic).
  3. Calendar events — list events for <date> in the local TZ.
  4. GitHub activity — .claude/skills/daily-recap/github-events.sh <date>. The script honors TZ, EXCLUDE_REPOS (comma-separated), and WORKDAY_START_HOUR env vars. Output:
    HH:MM | EventType | repo | action=... | ref=... | pr=... | title=... | commits=[...]
  5. Granola meetings — mcp__*Granola*list_meetings for the target date. Use Granola times when both calendar and Granola exist (Granola has actual durations).
  6. Claude Code sessions — .claude/skills/daily-recap/claude-sessions.sh <date> for session summaries; add --prompts for substantive user prompts. The script auto-derives the project scope from $PWD.

Slack timestamp accuracy

Slack search results return opaque message IDs, not real timestamps. To get accurate times:

  1. Collect the unique channel/DM IDs that had activity from the search results.
  2. For each active channel, use slack_read_channel (or equivalent) with oldest/latest set to the target date's Unix-timestamp range.
  3. Use those timestamps as the source of truth.

Step 2 — Enrich GitHub

For each unique PR number found, batch-fetch titles and diff sizes:

.claude/skills/daily-recap/pr-info.sh <owner/repo> <pr> [<owner/repo> <pr> ...]

Output: owner/repo#<pr> | <title> | +<additions>/-<deletions>. Use diff size to estimate review time (small <100 ≈ 5 min, medium 100–500 ≈ 15 min, large >500 ≈ 30 min). Batch into one invocation to avoid per-call permission prompts.

Step 3 — Build the timeline

Combine sources into one chronological timeline. For each entry:

  • Time (local TZ)
  • Source (Slack channel/DM, Calendar, GitHub repo, Granola, Claude project)
  • What happened — concise

Group related Slack messages into thread-level entries (one entry per thread, not per message). Merge corroborating signals (a GitHub push + Claude session for the same project = one entry).

For Claude Code sessions, one entry per distinct working window:

  • Time range (from first/last substantive prompt in the window)
  • Project (short name, not the slug)
  • What was worked on (summarized from prompts, not a prompt-by-prompt dump)

Scope: work, not personal chatter

Filter out personal/social content even if it flowed through Slack/calendar (after-hours plans, restaurant logistics, etc.). Keep "company event happened" at calendar-level only. Mixed work+personal threads → summarize the work portion only.

Step 4 — Summary

End with a Summary section. Stats first, then takeaways.

Stats

Each stat on its own line; bulleted detail beneath where applicable:

  • PRs Merged — count + bullets (repo, PR number, title)
  • PRs Opened — count + bullets
  • PRs Reviewed — count + bullets
  • Commits — count, e.g., "6 commits across 2 repos" (excluding EXCLUDE_REPOS)
  • Meetings — count + total duration + bulleted list (time, title)
  • Code Review — estimated minutes from PR diff sizes (~5/15/30 by size)
  • Slack — channels/DMs active in
  • Lunch — detect 30–60 min activity gap typically 12:00–13:30; show range or "No obvious lunch break"
  • Focus Time — blocks of 30+ min with no Slack and no meetings, work-hours only (infer end-of-workday from activity pattern; fall back to 18:00 if no signal). Exclude lunch. Sum and list specific blocks. If a block overlaps a Claude session, annotate the project worked on.
  • Longest Focus Block — the single longest uninterrupted focus block.
  • Evening Work (optional, informational) — substantive post-dinner work in one line; not counted as Focus Time.
  • Claude Sessions — count of distinct sessions with substantive activity + projects touched.

Takeaways

2–3 sentences covering key discussions/decisions and main themes of the day.

Step 5 — Stand-Up Block

Generate a stand-up-ready block in Daily Notes/-style format. Place it first in the output file (before Timeline and Summary) — it's the most-referenced part.

  • Yesterday = accomplishments from the previous working day (target date − 1). If target is Monday, label as Friday instead. To fill: read prior recap at _agent/outputs/recaps/<prev>.md; otherwise leave a single empty bullet.
  • Today = accomplishments on the target date itself, drawn from this recap's takeaways + key timeline threads.

Format:

# Stand Up
**Yesterday**
- <terse bullet>

**Today**
- <terse bullet>

Rules:

  • Terse, action-phrased, no trailing punctuation. Match the style of any existing stand-ups in the vault.
  • 1 to ~10 bullets. Group related work where natural ("Reviews — A, B, C"); don't force compression if the day had distinct threads.
  • PR / TASK references must include a short subject, not just IDs. PR#8343 rocket-engine expression library ✓ — PR#8343 ✗.
  • No meeting-by-meeting detail or Slack chatter — anything that wouldn't belong in a standup gets dropped.

Step 6 — Save

Write to _agent/outputs/recaps/<YYYY-MM-DD> - <DayOfWeek> - Recap.md (e.g., 2026-05-06 - Wednesday - Recap.md). Create _agent/outputs/recaps/ if missing.

If the file exists, read it first, merge in new entries, and preserve any manual edits. Don't duplicate.

Tag the file with source: agent in frontmatter per _agent/AGENTS.md.

Output format

Markdown with ### for time blocks, --- between sections, bold for channel/repo names. Order in the file:

  1. # <YYYY-MM-DD> - <DayOfWeek> - Recap header
  2. # Stand Up block
  3. ## Timeline (Morning / Lunch / Afternoon / Evening)
  4. ## Summary (Stats + Takeaways)

Caveats

  • Claude session file span ≠ working window. A session file can stretch across hours while real activity is clustered. Use --prompts timestamps for the truth.
  • Cross-day sessions are handled by the working-day window (03:00 → 03:00 next day).
  • Sibling project dirs like claude-b, claude-investigate are ephemeral scratch workspaces — name them by what was worked on (from prompts), not the dir name.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.