The repo that cleans up the fleet. Detects duplicates, plans merges, assembles collaborations.
912+ repos. Multiple agents building overlapping things. 5 FLUX runtimes. 6 holodeck implementations. 3 fleet infrastructure repos that do similar things. The fleet needs a dedicated agent whose job is to make order from abundance.
graph LR
Scan[Scan Fleet] --> Categorize[Categorize Repos]
Categorize --> Detect[Detect Duplicates]
Detect --> Plan[Plan Merges]
Plan --> Execute[Create PRs]
Execute --> Verify[Run Dockside]
Verify --> Report[Fleet Report]
style Scan fill:#bbf
style Detect fill:#fbb
style Execute fill:#bfb
- Scan — enumerate all repos, extract languages, files, signatures
- Categorize — group repos by functional domain (flux, holodeck, fleet-infra, etc.)
- Detect — find repos with overlapping functionality
- Plan — generate merge strategies (which repo wins, what gets absorbed)
- Execute — create PRs, move files, update references
- Verify — run dockside exam on merged results
| Group | Members | Status |
|---|---|---|
| FLUX runtimes | 8 repos (Python, C, Go, Rust, Zig, JS, WASM, Java) | Scanning |
| Holodeck | 7 repos (Rust, C, Python, Go, Zig, CUDA, TUI) | Scanning |
| Fleet infra | 5 repos (API, keeper, mechanic, GitHub app, orchestrator) | Scanning |
| Agent standards | 4 repos (standard, dockside, bootcamp, z-bootcamp) | Scanning |
| I2I/Comms | 4 repos (I2I, bottles, baton, mesosynchronous) | Scanning |
python3 engine.pyOutputs refactor-report.json with duplicate analysis and merge proposals.
The fleet is organic. Repos grow, fork, overlap, and sometimes become redundant. The Shipwright doesn't judge — it organizes. Every repo contributed something. The merge preserves that contribution while reducing noise for the next agent who needs to find what they're looking for.
Part of the Cocapn Fleet. The reef builds itself.