Musketeer logo Musketeer

What the CLI Does

The Musketeer CLI is the trio execution harness for SMALL-governed workspaces. It adds role-separated execution, packets, verdicts, and execution logs on top of SMALL-owned canonical state.

Core capabilities

  • Workspace initialization - bootstraps .small/ if missing, creates .musketeer/ execution layer
  • Role-separated execution - originator, examiner, and executor workflows with explicit handoffs
  • Packet generation - reads canonical state from .small/, assembles bounded role context packets in .musketeer/packets/
  • Execution logging - writes execution-log.yml in .musketeer/runs/<replayId>/
  • Verdict gate - auditor verdicts written to .musketeer/verdicts/<replayId>.verdict.yml
  • Dual validation - check validates SMALL state integrity and Musketeer execution state separately
  • Migration - migrate converts legacy workspaces to SMALL-native layout

What the CLI produces

Musketeer produces execution-layer artifacts in .musketeer/:

.musketeer/
  musketeer.yml                          # workspace config
  packets/                               # role context packets
  verdicts/
    <replayId>.verdict.yml               # auditor verdicts
  runs/
    <replayId>/
      execution-log.yml                  # execution logs

Canonical state (intent, constraints, plan, progress, handoff) lives in .small/ and is owned by SMALL. Musketeer reads from it but does not write to it.

Design principles

  1. Canonical state is owned by SMALL - intent, constraints, plan, progress, and handoff live in .small/
  2. Musketeer adds the execution layer - packets, verdicts, and execution logs live in .musketeer/
  3. Roles are separated - the CLI tracks which role produced which state
  4. Handoffs are recorded - transitions between roles are captured on disk

Local-first operation

The CLI requires no external services. Everything runs on your machine:

  • No accounts or subscriptions
  • No network connectivity for core operation
  • No cloud storage
  • Your data stays local

Use cases

  • Structured handoffs between different AI models
  • Audit trails of intent, constraints, and execution
  • Local development without cloud dependencies
  • State that can be inspected and version-controlled

See the usage documentation for specific commands.