Every order in a voyage knows which role should run it. Crews bind to model tiers automatically. Quality gates run between roles, not as a CI afterthought. Maestro Hub is the dashboard you run it all from.
Everything below runs from a terminal. It also runs from Maestro Hub: one dashboard across every registered project, served by the pip package on your own machine. No Docker, no Node, no Maestro cloud.
Live voyages, orders, crews, queue depth, token usage and the streaming lifecycle log, across every project at once rather than one terminal at a time.
Approve a plan, request changes or decline it. Retry a stranded order, hold one back from dispatch, or pause the fleet so running crews finish and no new work starts.
Create a voyage, file an order against it, and edit one that has not started yet. The Hub writes through the same code the CLI does, so the artefact is identical either way.
The Hub does not do everything the CLI does, and it says so: merging, tagging, publishing and workspace migration stay command-line and human. Its documentation publishes the full boundary rather than leaving you to discover a missing button.
The Lead sits at the top of the chain of command, and it dispatches and oversees the six specialised squad roles below. Each role has a canonical code used in order files and config. The active persona overlays a themed display name, but the code is stable across all personas.
Orchestrator. The local interactive Claude session: talks to you, dispatches every squad, and tracks voyage progress. Owns the chain of command above all six squad roles. Opus 5.
Dispatches · sequences · oversees
Plans the voyage. Produces a written plan in .mso/plans/ before any other crew runs. Opus 5.
Implements features. Reads the plan, writes the code, runs the tests it touched. Sonnet 5.
Tests + validates acceptance criteria. Refuses to sign off if any AC fails. Sonnet 5.
Documentation. Updates README, CLI reference, release notes. Sonnet 5.
Security review. Read-only. Runs OWASP/CWE pattern scans. Reports findings, never fixes. Haiku 4.5.
Integration gate. Opens the PR, triages CI, resolves review-bot threads, then signs off READY or NOT READY. Never merges: that call stays yours. Opus 5.
Watch a real order advance through its role sequence. Each crew hands off only when its acceptance criteria pass. Add REL and the Releaser triages CI and the review threads, then recommends READY or NOT READY. Merging stays a human decision.
Canonical codes are fixed. Each built-in persona substitutes its own display names in prompts and CLI output. The mechanics underneath are identical.
| Code | Default | ⚓ Nautical | 🔱 Navy SEALs | ⚡ Storm Troopers | 🌿 Game Rangers |
|---|---|---|---|---|---|
| PLN | Planner | Navigator | Recon | Strategist | Scout |
| BLD | Builder | Shipwright | Operator | Engineer | Tracker |
| TST | Tester | Bosun | Range Safety | Inspector | Warden |
| DOC | Documenter | Scribe | Debriefer | Scribe | Chronicler |
| SEC | Security | Lookout | Intel | Counter-intel | Ranger |
| REL | Release | Coxswain | Extraction | Commander | Conservator |
| LEAD | Lead | Quartermaster | Command | Director | Chief Ranger |
Set your persona with mso settings set persona <id>, or export MAESTRO_PERSONA to override it for a single process. See the persona picker for a full preview.
A real-time dashboard across every registered project: voyages, orders, crews, queue depth, token usage and the review queue. Runs from the pip package on your own machine, with no Docker and no Node required. Team and Enterprise.
A per-workspace background agent that keeps watching the fleet after your terminal closes. It runs the mso doctor patrol on a cadence and handles two reversible housekeeping actions on its own (archiving merged voyages, sweeping the artefact plane), always logging what it did and why. Pro and above.
Source control, CI and issue tracking resolve through a provider seam: GitHub or Bitbucket, GitHub Actions or Jenkins, GitHub Issues or Jira. An unconfigured provider is reported as a named skip, never a silent pass.
A read-only patrol across dispatcher, crews, orders, voyages and machine hygiene. It reports and proposes a command; it never mutates. Works with the dispatcher stopped.
mso pause drains the fleet: no new work is claimed, running crews finish, then the dispatcher exits cleanly. mso resume picks up where it left off, with no cleanup and no retries.
mso settings list / get / set resolves every key the same way for the CLI and the Hub, reporting the effective value, its source and its precedence. Per workspace or per device.
Discrete tasks (orders) grouped into sprints (voyages). dependsOn chains enforce order. One voyage = one branch = one PR.
Three validation types: file-exists, manual judgement, shell command exit code. Tester validates; Releaser enforces.
AWS keys, GitHub PATs, OpenAI / Anthropic / Slack tokens, JWTs, RSA private keys. Blocks the tool call before the secret hits a commit.
SQL injection, XSS, path traversal, hardcoded credentials, insecure deserialization, command injection, weak crypto. Findings written to .mso/reports/security/.
Per-server risk levels and per-role permissions. The Lead refuses to launch crews that request unregistered or denied servers.
Hash-chained JSONL of every privileged action. Optional external anchoring to S3 Object Lock / Azure Immutable Blob.
Mobile control. You receive crew questions on Slack and reply; the crew picks up the answer and continues.
Full workspace state to a single ZIP. Restore restores orders, voyages, plans and reports: everything except runtime state.