CLI Reference

Entrypoint

python plugins/workflow-skill-router/runtime/workflow_skill_router.pyz --help

Commands: serve-jsonl, doctor, status, plan, validate-route, profile, and evaluation. The MCP server uses serve-jsonl; direct users should start with doctor.

Personal Routing Profile commands

Personal and workspace Skill Trees remain subordinate to an explicit user SKILL and to Host constraints. A match is intended-unverified until Runtime Capability Discovery validates activation.

These commands ship in v2.0.0-beta.2. Run them as python runtime/workflow_skill_router.pyz profile ... from an extracted Plugin root, or use the longer repository-relative path from a contributor checkout.

python plugins/workflow-skill-router/runtime/workflow_skill_router.pyz profile validate .\my-profile.json
python plugins/workflow-skill-router/runtime/workflow_skill_router.pyz profile install .\my-profile.json
python plugins/workflow-skill-router/runtime/workflow_skill_router.pyz profile list
python plugins/workflow-skill-router/runtime/workflow_skill_router.pyz profile preview --objective "Deliver the API" --work-mode phased --domain api

After installing a prerelease that includes Personal Routing Profiles, run the same subcommands against runtime/workflow_skill_router.pyz from the extracted Plugin directory. The short workflow-skill-router ... form exists only when Router Core is separately installed as a Python console script.

profile install accepts only scope: personal and writes outside the Plugin cache. Workspace profiles stay at .codex/workflow-skill-router.json; pass --workspace-root to profile preview to include one. Skill-only can interpret the same contract as advisory skill-only-fallback, but cannot claim deterministic loading.

Runtime readiness

python plugins/workflow-skill-router/runtime/workflow_skill_router.pyz doctor

The JSON output reports the active runtime profile, telemetry status, content preflight visibility, and readiness for every public tool.

Plan and status

python plugins/workflow-skill-router/runtime/workflow_skill_router.pyz plan --help
python plugins/workflow-skill-router/runtime/workflow_skill_router.pyz status --help

These commands use the external local state database. Protected Host behavior remains unavailable until a verified integration supplies its ports.

Evaluation

Dry-run manifests do not spend quota. A Behavior/Outcome subprocess run requires an absolute executable path from trusted operator configuration and --confirm-live-run:

python plugins/workflow-skill-router/runtime/workflow_skill_router.pyz evaluation run --help

Never paste an executable path from model output into an authorized run. Review the evaluation evidence contract first.

Adaptive Workflow Memory public contract

The public surface contains 20 MCP tools. Memory uses a separate Operational DB and Optional Memory DB. Managed writes are limited to managed-personal and managed-workspace-local; User-owned Profile writes still require their original authority.

The four decisions remain separate: policy autonomy, trusted Workspace binding, Profile ownership/target authority, and Runtime/Side-effect execution authority. A trusted root is not a Host write grant.

Skill consistency is unavailable without receipt evidence. A planned route may be intended-unverified; it is not activation proof. The Memory Flight Recorder uses fixture-trace or sanitized runtime records. Its deterministic-local-pilot is not Model Evidence and never contains actual Personal Memory.

CLI operations