The MCP server gives your agent the model. This gives it the method.
An agent that can read a conceptual schema still has to know how to build one. Modelling is a procedure with an order to it — verbalize examples first, constrain second, never assert a rule nobody confirmed — and an agent that does not know the order produces a diagram that looks like ORM and says nothing checkable.
The pack is ten slash commands and two skills distilled from Fact-Based Agents. They carry the procedure, the constraint family, the validator codes and three worked schemas, so the agent follows the method rather than improvising one.
Factum$ factum skills install
Where should the Factum skills go?
1) Claude Code ~/.claude all your projects
2) Claude Code ./.claude this project only
3) Cursor ~/.cursor all your projects
4) Cursor ./.cursor this project only
5) Codex CLI ~/.codex all your projects
6) Codex CLI ./.codex this project only
7) OpenCode ~/.config/opencode all your projects
8) OpenCode ./.opencode this project only
Each one is a procedure with a stopping point, not a prompt. /orm-model works
Halpin's seven steps and stops at the end of each for a yes or no, because a constraint you did
not confirm is a constraint the model should not carry.
| Command | What it runs |
|---|---|
/orm-model | Build a model from real examples, following the seven-step design procedure |
/orm-derive | Draft a first model from a CSV, a SQL schema, a JSON payload or an existing codebase |
/orm-review | Review a model against the validator, the anti-patterns and the pre-commit checklist |
/orm-fix | Repair validator errors, one code at a time |
/orm-verbalize | Turn a model into FORML sentences for a domain expert, an agent, or a review |
/orm-diagram | Lay out and render the diagram, or explain an existing one |
/orm-map | Generate the relational and property-graph schemas, and read the mapping notes |
/orm-evolve | Plan a schema change: diff the model, measure drift, stage the migration |
/orm-memory | Design an agent's memory, promise graph or audit log as a conceptual schema |
/orm-explain | Explain an ORM 2, FORML or fact-based modelling concept with a worked example |
A skill loads itself when the work calls for it — you do not have to remember it exists.
Each is a SKILL.md with its reference material and example models beside it, so the
agent reads the detail it needs rather than all of it.
factum-ormThe notation, the constraint family, the design procedure, the validator codes and the
mapping rules — with the file format and eleven example models to take apart. Loads
whenever a task mentions ORM, FORML, elementary facts or .orm.json.
factum-agentic-memoryThree finished conceptual schemas and the reasoning behind each decision: the six memory layers, a Promise Theory promise graph for multi-agent coordination, and a tamper-evident audit log. Loads when the work is agent memory, provenance or coordination.
Every one of these agents reads the same SKILL.md layout, so nothing is converted on
the way in. What differs is the directory, and whether the agent reads slash commands at all.
| Agent | Skills install to | Slash commands |
|---|---|---|
| Claude Code | ~/.claude/skills · ./.claude/skills | ✓ yes |
| Cursor | ~/.cursor/skills · ./.cursor/skills | ✓ yes |
| Codex CLI | ~/.codex/skills · ./.codex/skills | – skills only |
| OpenCode | ~/.config/opencode/skills · ./.opencode/skills | – skills only |
Codex and OpenCode read the skills but not the pack's slash commands, so those are left out rather than written into a directory nothing reads — the same procedures are in the skills either way. Nothing already installed is replaced unless you ask:
factum skills list # what is in the pack
factum skills install --target cursor --local
factum skills install --dir ~/somewhere/else
factum skills install --force # replace what is there
factum skills install --dry-run # say what would happenfactum-mcp gives the agent eight tools over a .orm.json file: read
it, verbalize it, validate it, map it, diff it, and write it back through a guarded save.
That is ground truth about your domain, and it is what stops the agent inventing semantics from your table names.
The skills give it ORM itself: what an elementary fact is, which constraint says what, the order to apply them in, and what the validator means when it refuses.
Either works alone. Together the agent knows both the domain and how to model it, which is when it stops asking you to explain uniqueness constraints and starts drawing them.
Install the CLI and run one command. Everything runs locally, and nothing here calls a language model — the skills are instructions for the agent you already have.
npm install -g factum-orm
factum skills install