Skip to main content

​
Core Routines

Core routines are the backbone of the EvoNexus daily loop. They ship with the repo, are hardcoded in scheduler.py, and run automatically. Unlike custom routines (which live in ADWs/routines/custom/ and are configured via config/routines.yaml), core routines require zero configuration. There are 6 core routines. Together they form a closed loop: orient in the morning, work during the day, consolidate at night, backup data, sync memory, and review weekly.

​
Daily Backup

Why it matters: Gitignored data (memory, configs, logs, customizations, dashboard DB) is lost on clean installs or machine changes. This routine exports all user data as a ZIP, enabling portability without breaking git pull updates. If S3 is configured, it also uploads the backup to the cloud. Execution: Pure Python via run_script() (no AI, no tokens) Timeout: 5 minutes

​
What it does

  1. Collects all gitignored files using git ls-files --others --ignored --exclude-standard
  2. Filters out heavy reconstructible dirs (node_modules/, .venv/, __pycache__/, dist/)
  3. Creates a ZIP with manifest.json (version, timestamp, hostname, file list)
  4. Saves to backups/evonexus-backup-YYYYMMDD-HHMMSS.zip
  5. If BACKUP_S3_BUCKET env var is set, uploads the ZIP to S3

​
Output

A ZIP file in backups/ containing all workspace user data. Restorable via make restore FILE=<path>.

​
CLI

​
Good Morning

Why it matters: Without a structured start, the day drifts. This routine scans your calendar, emails, tasks, and recent work to give you a clear picture of where you left off and what needs attention today. Skill: prod-good-morning Timeout: 10 minutes

​
What it does

  1. Reads CLAUDE.md (master context) and the last 3 session logs from workspace/daily-logs/
  2. Reads active project overviews
  3. Gathers live data:
    • Today’s calendar via Google Calendar MCP
    • Unread emails via Gmail MCP
    • Today’s tasks via Todoist
  4. Delivers a brief recap (pt-BR):
    • Recent work (2-4 bullets)
    • Open/mid-flight items
    • Today’s meetings and times
    • Attention-needed emails
    • Priority tasks
  5. Gives one clear recommendation on what to work on first
  6. Asks what you want to tackle

​
Output

A structured morning briefing displayed in the terminal. No file saved — this is an interactive session starter.

​
End of Day

Why it matters: Without consolidation, context is lost between sessions. This routine captures everything that happened — decisions, learnings, people updates, pending items — so tomorrow’s Good Morning has fresh context to work with. Skill: prod-end-of-day Timeout: 10 minutes

​
What it does

  1. Collects data from the entire day:
    • Agent memory updates from .claude/agent-memory/ (all agents)
    • ADW logs from ADWs/logs/YYYY-MM-DD.jsonl
    • Meeting summaries from workspace/meetings/summaries/
    • Tasks completed/created via Todoist
    • Git changes (git diff --stat, git log --oneline --since="today 00:00")
    • Current session conversation
  2. Consolidates into a daily log:
    • Decisions made and why
    • Patterns, feedback, and corrections
    • New context about people
    • Real pending items (not aspirational)

​
Output

A daily log saved to workspace/daily-logs/[C] YYYY-MM-DD-eod.md.

​
Memory Sync

Why it matters: Memory files drift out of date if not actively maintained. This routine reads recent daily logs, meetings, and git history to extract new facts and propagate them across the memory system, keeping it consistent and current. Execution: Direct prompt via run_claude() (not a skill) Timeout: 10 minutes

​
What it does

  1. Reads last 3 daily logs from workspace/daily-logs/
  2. Reads meeting summaries from the last 3 days
  3. Analyzes git log --oneline --since="3 days ago"
  4. Extracts and classifies:
    • Decisions —> project type memories
    • People context —> user type memories
    • Feedback/corrections —> feedback type memories
    • New terms/references —> reference type memories
  5. De-duplicates: updates existing memories instead of creating duplicates
  6. Ingest propagation: When updating a memory, finds and updates other memories that reference the same entity
  7. Updates MEMORY.md index and memory/index.md
  8. Appends to memory/log.md: [DATE] SYNC -- summary

​
Output

A report showing count of memories created/updated by type and cross-references propagated.

​
Memory Lint

Why it matters: Over time, memory accumulates contradictions, orphan files, stale claims, and coverage gaps. This weekly health check catches problems before they compound, keeping the knowledge base reliable. Execution: Direct prompt via run_claude() (not a skill) Timeout: 10 minutes Schedule: Every Sunday at 09:00

​
What it does

  1. Reads memory/index.md to get the full catalog
  2. Verifies every listed file exists and is not empty
  3. Detects orphan files (exist in memory/ but not in the index)
  4. Cross-reference checks:
    • People mentioned in projects should have a file in memory/people/
    • Projects mentioned in people files should have a file in memory/projects/
    • Glossary terms should point to valid entries
  5. Stale data check:
    • Flags Linear issues that may be closed
    • Flags people roles/companies that may have changed
    • Flags any claim with a date older than 60 days
  6. Contradiction check:
    • Compares CLAUDE.md hot cache with memory/glossary.md
    • Compares people info in CLAUDE.md vs memory/people/*.md
  7. Coverage gaps:
    • Active projects in CLAUDE.md without a memory/projects/ file
    • People in CLAUDE.md without a memory/people/ file
  8. Auto-fixes what it can:
    • Updates memory/index.md with orphan files
    • Fixes obvious contradictions (prefers the more recent source)
    • Adds missing cross-references
  9. Appends to memory/log.md: [DATE] LINT -- summary

​
Output

A report listing issues found, fixes applied automatically, and items needing manual review.

​
Weekly Review

Why it matters: Daily logs capture details, but the weekly review steps back to see the big picture — what actually got done, what’s carrying over, and what’s coming next week. It produces a formatted report you can reference or share. Execution: Direct prompt via run_claude() (not a skill) Timeout: 15 minutes

​
What it does

  1. Gathers the full week’s data:
    • Meetings synced via /int-sync-meetings
    • Tasks via /prod-review-todoist (completed, overdue, next week)
    • Next week’s agenda via Google Calendar
    • Memory consolidation from the week’s daily logs
  2. Generates reports in two formats:
    • HTML using template .claude/templates/html/weekly-review.html
    • Markdown using template .claude/templates/weekly-review.md

​
Output

Dual-format report saved to workspace/daily-logs/:
  • [C] YYYY-WXX-weekly-review.html
  • [C] YYYY-WXX-weekly-review.md

​
The Daily Loop

The core routines form a continuous feedback loop:
Each routine feeds the next. Good Morning reads what End of Day wrote. Memory Sync propagates what End of Day captured. Weekly Review consolidates what Memory Sync organized. Memory Lint ensures the whole system stays healthy.

​
Running Manually

All core routines can be run at any time — they are not restricted to their scheduled times.