wizard

name	wizard
description	Goal-first amq-squad preparation and launch wizard. Use it to turn a request or source into reviewed coordination artifacts, prove exact roster and bootstrap readiness, and present the separate default-No launch gate.

Skill version: 2.22.0 - Start the first response by stating the loaded identity as amq-squad skill v2.22.0 before status or analysis.

amq-squad:wizard

Use this operator-facing skill before a new squad launches. It owns goal intake, artifact preparation, readiness, and final launch preview. It never treats a syntactically valid launch command as proof that the team is ready.

Immutable stage contract

The stages are goal, brief, rules, roles, profile, readiness, and launch. Every stage defaults to read-only. A later stage consumes the accepted output of the earlier stage without silently changing its goal, namespace, rosters, topology, role contracts, or tool policy.

Preparation and launch are separate approvals:

1. Render the proposal and exact project-local mutations.
2. Obtain explicit preparation approval before writing coordination artifacts.
3. Run readiness against the written artifacts and generated bootstrap preview.
4. Present a separate default-No launch confirmation for exactly the displayed initial roster.

Preparation never launches panes. Launch never repairs or rewrites accepted artifacts.

For a non-interactive operator or CI flow, preserve the same four stages and the same argv identity:

amq-squad run start --project P --profile R --session S --roles cto,qa \
  --lead cto --launch-shape working-team-together --goal "..." --prepare-plan
# Default No: repeat only after the operator accepts the rendered proposal.
amq-squad run start --project P --profile R --session S --roles cto,qa \
  --lead cto --launch-shape working-team-together --goal "..." --prepare
amq-squad run start --project P --profile R --session S \
  --launch-shape working-team-together --readiness-json
# Separate default-No launch approval; use the exact accepted binding values.
amq-squad run start --project P --profile R --session S \
  --launch-shape working-team-together --goal "..." \
  --goal-source operator_goal --goal-digest 'sha256:<accepted-digest>' --go

--prepare-plan, --prepare, and --go are not aliases for one another. Never tell an operator to jump from a generic preview directly to --go.

Goal binding

A launch requires an actionable goal binding for the visible lead. Show its source, exact profile/session namespace, text or bounded digest, delivery method, and validation status.

• Explicit goal text is reviewed verbatim.
• When goal text is blank and the exact namespace already has a real accepted non-stub brief, derive the deterministic directive Execute the accepted brief for namespace <profile>/<session> at <path>. and require operator acceptance. Never rewrite the brief.
• Blank goal plus a missing or generated-stub brief fails readiness. It must never produce a live prompt_goal_missing run.

Composition proposal

Render separately:

• initial launch roster: count, names, binary, model, effort, intent, mutation authority, and effective tool profile;
• staged-later roster: count, names, join condition, and spawn-gate requirement;
• launch shape: explicitly working-team-together or lead-only-staged.

Orchestration or a visible lead never implies lead-only launch. Existing profiles without an accepted launch shape are legacy/unspecified and require operator confirmation.

Readiness rows

Emit machine-readable rows with ready, missing, stub, generic, stale, or drifted, plus evidence and deterministic fix/preview actions, for:

• accepted brief and goal binding;
• team rules and operator/orchestration policy;
• every initial and staged role contract;
• profile membership, binary/model/effort/execution/tool policy;
• initial versus staged roster equality;
• one generated bootstrap row for every initial member and no staged-only role;
• binary/skill, AMQ, pointer, and launch-capability diagnostics.

Readiness fails closed when any required row is not ready, profile/bootstrap membership differs from the accepted initial roster, or the goal binding is not verified. Runtime terminal capability data is consumed through the CLI-owned diagnostic contract; the wizard does not infer a backend.

Tool policy

For multi-agent teams recommend a broad lead and the smallest sufficient worker profile (minimal, coding, browser, data, or explicit full). Show the effective policy and source for each member. Claude settings overlays and Codex profiles must be materialized before the binary starts. Never silently remove a capability the operator explicitly requested.

Invocation

• Full flow: wizard <request>.
• One stage: stage goal|brief|rules|roles|profile|readiness|launch <request>.
• Canonical binary UI: amq-squad wizard --project P --profile R --session S.

After launch, the visible lead uses amq-squad:orchestrator; direct operations and diagnostics use amq-squad:cli.