🔀 Decision: Wipe and Rebuild Paperclip Org
Date: 2026-08-03 Status: Proposed (awaiting Pat’s final go) Decided by: Pat + J.A.R.V.I.S.
Context
Paperclip agents have been running for weeks but producing unreliable results:
- 5 Huible build issues marked “done” but code is stranded in workspace with no git
- Multiple agents in error state (CEO, InvestInMe PM, R&D Lead) — model configs pointing to unavailable models
- Project Builder has been erroring since July 13
- Agent heartbeats were burning tokens even when idle with nothing to do
- Cross-company access issues blocking work
- Agents writing Cloudflare debugging notes into Clinical Advisor issues (context contamination)
Decision
Nuke existing Paperclip agents and rebuild from scratch with:
- Proper agent instructions (pointing at vault for context)
- Correct model configurations (all using available models per Model Strategy)
- Git-first enforcement (agents can’t mark done without push)
- Clear department boundaries (no cross-project contamination)
- Heartbeats only when there’s active work
Reasoning
The current org accumulated cruft over weeks of experimentation. Fixing each agent individually would take longer than a clean rebuild. The vault now holds all the knowledge — agents just need to be pointed at it.
Consequences
- All existing Paperclip agents will be deleted or archived
- New agents created with vault-aware instructions
- Huible code will be rescued into git repo first (preserving work)
- Sprint cron jobs paused during rebuild
- Clean org chart: one PM + one Tech Lead per project, reporting to CEO