Skip to content
Content modellingCMMF 1.2.0

Work with CMMF Skills

Build and install complete development packages, then try a modelling workflow with explicit source and validation boundaries.

Skills give a coding agent CMMF-specific instructions and local tools. Begin with a content brief or an existing model set; use the agent to propose and review changes, then inspect the files and validation evidence yourself.

Skill Job
cmmf-discover Turn workshop notes into a modelling brief
cmmf-author Create model files from an agreed brief
cmmf-validate Inspect structure and whole-model errors
cmmf-review Review modelling choices and editorial usability
cmmf-document Produce a catalogue and relationship documentation
cmmf-evolve Assess a proposed model change and its impact

The host supplies reasoning, authentication and file access. The local helpers do not require another LLM provider key. Host permissions, subscriptions and usage limits still apply.

Obtain the Skills repository, review its current README and select the revision you intend to evaluate. Use its reviewed Node/npm requirements; the documented helper runtime needs Node 22 or newer.

Run these commands inside that checkout:

Terminal window
npm ci
npm run check
npm run skills:package
npm run skills:verify

Packaging creates self-contained directories under release/skills/, plus release/cmmf-skills.tar.gz and a file-hash manifest. Dependencies and their licences are included in the packages; installed helpers do not need a separate npm installation.

This site does not offer a Skills binary download: its evidence includes working-tree documentation at base commit 53cf3928, rather than a reproducible published suite release. Review the source checkout and its generated manifest before installation.

Copy the entire packaged skill, including references, licences and helpers. Copying only SKILL.md or the development source directory leaves out the packaged runtime.

Host Project directory Personal directory
Codex .agents/skills/ ~/.agents/skills/
Claude Code .claude/skills/ ~/.claude/skills/
GitHub Copilot .github/skills/ ~/.copilot/skills/

These local directory conventions are documented by OpenAI, Anthropic and GitHub, checked on 15 October 2026. Check the instructions for your specific host and environment when installing.

For example, from the Skills checkout, copy Author into a model project’s Codex directory:

Terminal window
mkdir -p /path/to/model-project/.agents/skills
cp -R release/skills/cmmf-author /path/to/model-project/.agents/skills/

Replace the example path with your project’s location. Inspect an existing destination before copying. Preserve local customisations when updating; uninstall by removing only the installed directory you own. Use the host’s skill selector to confirm discovery before expecting a workflow to run.

Extract the Customer Story project into a working directory. Install cmmf-review and cmmf-validate there, then ask:

Use cmmf-review to assess this Customer Story set against its README brief. Explain the linked customer and embedded quote decisions. Propose improvements without changing files.

Inspect the proposed changes. When authorising an edit, state its scope and ask for complete-set validation, the source version/revision, changed files and remaining uncertainty. Run the project’s npm run validate yourself after installing its dependencies.

New projects default to the latest CMMF source. Existing projects use the manifest-declared version. An explicit historical request needs matching schema, specification and revision evidence. Offline cache reuse does not establish that “latest” is still fresh.

The documented helper semantics recognise legacy 1, 1.1.0 and 1.2.0; unsupported versions must be reported rather than silently upgraded. Keep source caches outside model directories.

Helpers inspect, resolve sources, validate, compare and document models. They do not validate actual content instances, transfer content, run live CMS migrations or provide CMS import/export adapters. Review those implementation decisions separately. See the validation guide and tool evidence.