ADR-0003 — OpenClaw permanently rejected for any JB project¶
Status: Accepted (Final — not open for reconsideration) Date: 2026-04-07 Decision-maker: João Brazão
Context¶
OpenClaw was evaluated for workflow automation across multiple projects. After deployment, the ClawHavoc disclosure revealed: - ~21,000 OpenClaw instances exposed on the public internet - Pattern of misconfiguration at scale, not a single bad deployment - Direct path to health-data exfiltration in RGPD Article 9 contexts
The risk is not theoretical — it is empirically demonstrated by mass exploitation.
Decision¶
OpenClaw is permanently removed from all JB infrastructure. The OpenClaw container was deleted on 2026-04-07, freeing 899 MB of RAM. No project may reintroduce it, regardless of perceived containment.
This decision is final. It does not require periodic re-evaluation.
Alternatives¶
| Need previously met by OpenClaw | Replacement |
|---|---|
| Workflow automation | Paperclip skills in jb-skills-gateway |
| HTTP webhooks | FastAPI microservices |
| Cron-based jobs | Plain cron + skills |
| Email triage | triage-bot + triage-dashboard containers |
Consequences¶
Positive - ~900 MB RAM recovered - RGPD Article 9 audit trail simplified — one less platform to justify - Single workflow runtime (Paperclip skills) instead of two
Negative - No-code visual workflow editing is gone - Skills must be written by humans (this is also positive in context — see adr/0001-no-cloud-no-github.md)
Related¶
- n8n was also removed (2026-04-07) for the intelligence-engine use case it covered, replaced by Paperclip skills.