Workflow Orchestration
MultiModel Dev OS v2.5.0 features a built-in Workflow Runner designed to orchestrate codebase diagnostic pipelines safely.
1. CLI Commands
Workflows are registered in .ai/registries/workflows.yaml and can be executed via the following subcommands:
List Registered Workflows
Prints name, risk, and summary of all active workflows:
npx multimodel-dev-os workflow listShow Workflow Specifications
Displays details, memory write permissions, code modification permissions, and individual logical steps:
npx multimodel-dev-os workflow show release-checkWhen a workflow declares optional Skill OS metadata, workflow show also displays the referenced skills, prompts, permissions, guardrails, and required context files. This metadata is read-only and does not change execution behavior.
Plan Workflow Execution (Dry-Run)
Prints the execution sequence, command lists, and expected outputs without executing any logic:
npx multimodel-dev-os workflow plan repo-healthRun Workflow
Executes the workflow steps sequentially:
npx multimodel-dev-os workflow run repo-healthTIP
You can also run, plan, and list workflows interactively through the TUI Dashboard under the Quality Gates & Diagnostics or Memory & Intelligence menus.
2. Standard Built-in Workflows
MultiModel Dev OS ships a bundled workflow registry with common diagnostic and development workflows:
repo-health(Low Risk): Scans framework signals, performs advisory doctor audits, and verifies file structures.memory-refresh(Medium Risk): Assesses memory differences and incremental refreshes.feedback-review(Low Risk): Lists active developer logs and compiles rules tolearning-rules.md.proposal-review(Low Risk): Inspects codebase proposals, checks status counts, and displays audit apply logs.release-check(Low Risk): Verifies codebase structures, executes release doctors, and runs package pack checks.
Some bundled workflows also include optional skill_os references that connect them to Skill OS metadata:
skill_os:
skills:
- release-governance
prompts:
- release-audit
permissions:
- git-push
guardrails:
- confirm-external-write
required_context:
- docs/release-state.mdThese references are declarative. Validation checks that IDs and paths are valid, but workflows do not execute skills, prompts, permissions, or guardrails.
Sprint F also adds draft-only business operator workflow examples:
operator-weekly-reviewoperator-content-briefoperator-project-pulse
These examples use skill_os metadata to point at generic operator skills and prompts. They do not call inboxes, calendars, drives, CRMs, analytics systems, ad platforms, or publishing systems.
For migration from workflow-only YAML to workflow plus skill_os metadata, see Skill OS Migration Guide. For a minimal YAML example, see Skill OS Examples.
3. Strict Safety Gates
The workflow engine enforces strict safety boundaries:
- No File Modifications: Allowed workflows only execute read-only checkups and incremental metadata updates (memory file compilation, feedback summaries).
- No Shell Execution: Shell command execution from
workflows.yamlis prohibited. Steps map directly to internal Javascript CLI functions. - No Autonomy: Any step requiring code changes (e.g. applying proposals) stops and outputs manual next-step instructions for the developer.
- Declarative Skill OS Links Only: Skill OS workflow metadata is inspected and validated only. It does not trigger automation or permission enforcement.
4. Bundled Registry Fallback
If the repository does not have a local .ai/registries/workflows.yaml registry file initialized yet, the workflow runner will automatically fall back to the bundled registry package templates and output a notice. This allows running read-only diagnostics prior to full project onboarding.
Curated Workflow Packs
You can extend the available workflows in your repository by installing specialized plugin packs from the Workflow Marketplace Catalog (e.g. npx multimodel-dev-os catalog install git-workflows --approved).
