Skip to content

SKILL.md

Latest commit

 

History

History
102 lines (71 loc) · 3.52 KB

SKILL.md

File metadata and controls

102 lines (71 loc) · 3.52 KB
Raw
name prod-end-of-day
description End-of-day consolidation — analyzes agent memory, ADW logs, meetings, tasks, and learnings to generate a complete daily log. Trigger when user says 'end of day', 'wrap up', 'done for today', 'goodnight', 'shutdown', or anything that signals finishing a work session.

End of Day — Daily Consolidation

End-of-day routine that consolidates everything that happened during the day: agent memory, ADW logs, meetings, tasks, and learnings.

Always respond in English.

Step 1 — Collect data for the day (silently)

Read all available sources without narrating each step:

1a. Agent memory

Read recent memory files from each agent in .claude/agent-memory/:

  • flux-finance/ — financial decisions of the day
  • atlas-project/ — project updates
  • kai-personal-assistant/ — if there is anything relevant
  • Any other agent that was used

1b. ADW logs

Read today's JSONL log at ADWs/logs/YYYY-MM-DD.jsonl to see which routines ran, duration, and status.

1c. Meetings of the day

Check workspace/meetings/summaries/ and workspace/meetings/fathom/ for today's synced meetings.

1d. Tasks

Run todoist today to see completed and pending tasks for the day.

1e. Git changes of the day

Run git diff --stat and git log --oneline --since="today 00:00" to see:

  • Files created, modified, or deleted today
  • Commits made (messages and authors)
  • Uncommitted changes (working tree)

This gives the real overview of what changed in the workspace — more accurate than reading the conversation.

1f. Current session

Review the current session conversation — what was discussed, decided, and done.

Step 2 — Consolidate learnings

Analyze everything that was collected and identify:

  • Decisions made — what was decided and why
  • Learnings — patterns, corrections, feedback that should be remembered
  • People — new context about team members
  • Real pending items — things that truly remain open (do not fabricate)

Step 3 — Save memory

If there are relevant decisions, learnings, or feedback, save to persistent memory in memory/ following the workspace memory system (see prod-memory-management).

Do not duplicate — check if similar memory already exists before creating.

Step 4 — Generate daily log

Read the template at .claude/templates/end-of-day-log.md and fill with the consolidated data.

Save to:

workspace/daily-logs/[C] YYYY-MM-DD.md

The log should include:

  • What was done (projects, tasks, meetings)
  • Files created or modified
  • ADW routines that ran (with status)
  • Pending items (only if real)
  • Where to resume tomorrow

Step 5 — Organize tasks

Review Todoist tasks directly (do NOT invoke /prod-review-todoist as a sub-skill — it sends a duplicate Telegram notification):

  • Run todoist today to list today's tasks
  • For each uncategorized or non-PT-BR task: rename/recategorize via todoist update
  • Report how many were organized

Step 6 — Confirm

Present a short summary:

## Day closed

**Log:** workspace/daily-logs/[C] YYYY-MM-DD.md
**ADW Routines:** {N} executed ({status})
**Tasks:** {completed}/{total} completed
**Memories:** {N} created/updated
**Learnings:** {N} recorded

**Tomorrow:** {sentence about where to resume}

Notify via Telegram

Upon completion, send a short summary via Telegram to the user:

  • Use the Telegram MCP: reply(chat_id="YOUR_CHAT_ID", text="...")
  • Format: emoji + routine name + main result (1-3 lines)
  • If the routine had no updates, send anyway with "no updates"