orchestrate-project-by-stepslisted
Install: claude install-skill Fractera/Agent-Engineering-Infrastructure
# orchestrate-project-by-steps
The **frozen process** for **projects** (automations / internal tools). You give a **proposed
node graph**; the engine decomposes it into a materialized queue of development sub-steps and
gates each on **spec completeness** — before any code is written.
> **READ FIRST — the automation-ontology glossary**
> (`CRUD-DOCS/workspace-standards/automation-ontology.md`, step 188-R): every automation is
> composed of EXACTLY twelve entities — Automation · Trigger · Hook · Condition · **Action** ·
> Router · Step | Integration · Channel · State | Run · Record. The graph you propose is a typed
> instance of that ontology and the engine VALIDATES it (schema v2, engine ≥0.8). An automation
> described outside the ontology is a defect. Note the terminology: **Action is the first-class
> entity** (a named outcome = a branch of steps, bound to its hook phrases); a work NODE is a
> `step` (the old node kind `"action"` is accepted only as a legacy alias of `step`).
This skill is **self-sufficient**: one `.mjs` (or, from D5, one MCP tool), no Hermes required —
a lone agent (even a project with only Codex) runs the whole process on its own.
## Why this exists (read once)
Content had a frozen process (`orchestrate-content-by-steps`); projects did not. Without deep,
materialize-first decomposition a project is "doomed to fail" (owner). From the content engine we
take **only the idea** — freeze the process: decompose and write the whole sub-step queue to disk
FIR