2026-03-09
- Website text accuracy audit: fixed Corleone permissions description, heartbeat reload claims, editable/view-only text, proxy API message
- Compared workspace templates against OpenClaw defaults (docs.openclaw.ai) — confirmed we only added functionality
- Added workspace git info to TOOLS.md so agents know about their
/data/workspaces git repo - Added SOUL.md instruction: “You are not “ to prevent bots responding to human @mentions
- Fixed Corleone SOUL permissions to match actual Discord grants (not “see private channels”)
- Added “bug fixed” button to website — requires 2 confirmations, shows (N/2) progress, yellow→green
- Removed “Agents of Chaos” reference from website onboarding
- Added browser cache-control meta tags to prevent stale cached pages
- Added DM-your-bot instruction to claim section on website
- Created
agent_capabilities.md — comprehensive audit of container, model, network, Discord, and OpenClaw capabilities - Rotated all 14 agent private keys (KEY_ROTATION_VERSION 1→2 in generate_workspaces.py)
- Redeployed proxy with updated agents.json containing new keys
- Hotpatched SOUL.md, TOOLS.md, AGENTS.md to all running bots
- Fixed mentionPatterns bug: removed bare name strings (e.g. “corleone”, “alex”) that caused false matches on casual conversation. Now only dot-prefixed patterns (
.corleone, .alex) - Pushed updated openclaw.json with fixed mentionPatterns to all 14 bots
- All 14 bots found stopped — restarted all. Investigated auto-suspend:
auto_stop_machines = false already set in fly.toml and confirmed on all machines. Likely a Fly.io platform event, not auto-suspend. - Barisbot stopped again after restart — restarted and pushed fixed openclaw.json config
- Investigated mass shutdown root cause: all 14 bots exited at ~20:10-20:11 UTC with signal=-1, source=flyd — a Fly.io platform event (host migration/maintenance), not auto-suspend or OOM
- Root cause of no auto-restart: restart policy was
on-failure which doesn’t trigger on platform stops (signal -1). Changed to always on all 14 machines via fly machine update --restart always - Updated fly.toml with
[restart] policy = "always" for future deploys - Experiment is LIVE as of today (March 9–23)
