flux-point-studios/substrate
Multi-repo primitive registry: each repo declares its reusable primitives in a substrate.json manifest, a zero-dependency generator compiles them into one SUBSTRATE.md graph — nodes, edges, orphans as dormant value, hubs to harden first — and staleness alarms surface at every session start so new work plugs into what exists instead of running parallel to it. A per-repo deliverables.json ledger rides the same check: anything built but not yet sent alarms at session start until the send is recorded. A prose-smell gate rides every Write/Edit: outward prose that carries AI-writing frames (contrastive negation, whether-you're triples, assistant artifacts, stock LLM vocabulary) is bounced back for a rewrite before it can ship. A memory lint rides the same session-start channel: memory files citing paths that no longer exist, index entries pointing at deleted memories, and unsent claims contradicted by the deliverables ledger all alarm until the agent repairs them - detection is mechanical, repair is never.
Regenerate SUBSTRATE.md from every repo's substrate.json manifest — exit 1 means a manifest problem that CI treats as red. Codex entry point for the substrate emit command.
Onboard a multi-repo workspace onto the substrate registry — scaffold substrate.config.json, seed a starter substrate.json in a named repo, generate SUBSTRATE.md, and install the sweep doctrine. Codex entry point for the substrate init command.
Scan a file (or all staged deliverables) for AI-writing smell before it ships Codex entry point for the substrate smell command.
Run the substrate registry check and interpret it — graph size, orphans as dormant value, hubs to harden, staleness alarms and coverage notes — then recommend the single next action. Codex entry point for the substrate status command.