daily-archivist

Uses the Agent Skills package format. Execution depends on the host tools and permissions; consult the frontmatter compatibility requirements where provided.

Daily Archivist

Audit knowledge quality. Fix what's safe. Flag what's not. Route findings to the right agent.

Inspired by the "daily-archivist" pattern: a scheduled agent that vets all knowledge files, fact-checks claims, fills gaps, cleans up quality — and leaves notes for other agents when it finds issues they should handle.

What It Does

1. Fact-check — Verify claims in memory files against actual state (file counts, project status, config values)
2. Gap-fill — Find undocumented decisions, orphan references, missing rationale
3. Clean up — Fix formatting, remove duplicates, merge overlapping entries, archive stale content
4. Route findings — Leave notes in memory/inbox/<agent>.md for the appropriate agent
5. Bug hunt — Spot broken references, missing files, config drift, stale data

Setup

1. Create the inbox directory

mkdir -p memory/inbox
# Create inbox files for each agent in your team
for agent in main coder reviewer researcher; do
  cat > memory/inbox/${agent}.md << 'EOF'
# Agent Inbox
<!-- Archivist leaves notes here. Read at session start, delete after acting. -->
EOF
done

2. Add inbox reading to your agent startup

Add this to your AGENTS.md or equivalent configuration:

## Every Session
1. Read memory files and feedback
2. **Check inbox:** Read `memory/inbox/<your-agent>.md` — act on pending items, delete after handling

3. Configure the verification script

Copy scripts/verify-facts.sh and customize the checks for your project. See comments in the script for examples.

4. Schedule the cron

# Example for OpenClaw
openclaw cron add \
  --name daily-archivist \
  --cron "0 3 * * *" \
  --agent janitor \
  --model claude-haiku-4-5 \
  --timeout-seconds 300 \
  --announce \
  --message 'Daily knowledge audit. Run verify-facts.sh, scan changes, fix what is safe, route findings to agent inboxes.'