UltraPlan — Council of Planners

Spawn 4 specialist agents with opposing mandates to stress-test a plan before committing. Each agent argues from their perspective. The orchestrator synthesizes into a unified plan with dissents noted.

When to Use

• Major architectural changes (embed nu-engine, switch LLM providers, new OS)
• Cross-cutting decisions that affect all fleet nodes
• When "just do it" risks bricking the fleet

Council Members

1. Architect

• Mandate: Elegant design, minimal complexity, clean interfaces
• Asks: "What's the simplest architecture that solves this?"
• Rejects: Over-engineering, premature abstraction, unnecessary layers

2. Operator

• Mandate: Reliability, observability, rollback capability
• Asks: "How do I debug this at 3am when it breaks?"
• Rejects: Black boxes, missing logs, no rollback path

3. Security

• Mandate: Supply chain integrity, least privilege, attack surface
• Asks: "What can go wrong? What's the blast radius?"
• Rejects: Trusted binaries, shared secrets, network exposure

4. Pragmatist

• Mandate: Ship it. Time-to-value. Incremental delivery.
• Asks: "What's the smallest useful increment we can deploy today?"
• Rejects: Big-bang migrations, analysis paralysis, perfection

Process

1. Brief: Orchestrator states the problem and constraints
2. Independent analysis: Each agent writes their perspective (parallel)
3. Debate: Cross-review — each agent critiques the others
4. Synthesis: Orchestrator merges into one plan with:
   • Consensus items (all agree)
   • Compromises (2-3 agree, dissent noted)
   • Blockers (any agent can veto with justification)
5. Output: Single plan document with phases, tasks, and risk register

Invocation

/ultraplan <topic>

The orchestrator spawns 4 agents, collects their analyses, runs the debate round, and produces the final plan.

Verification contract (auto-injected by proveryay)

Every agent dispatched through this skill receives the following appended to its prompt automatically:

<see ~/.claude/skills/proveryay/CLAIMS-TEMPLATE.md — injected verbatim>

This means every agent MUST emit a <claims> block when it makes any deployment/state-change assertion. The proveryay SubagentStop hook will verify each claim in parallel and inject structured verdicts into the dispatcher's next-turn context.

If the agent's work is purely planning / research / narrative (no state changes — typical for ultraplan), the agent may omit the <claims> block. Proveryay treats absence as silent (no noise).

Invocation example with template injection:

# Prompt injection: template auto-appended (idempotent)
PROMPT="${ORIGINAL_PROMPT}

$(cat ~/.claude/skills/proveryay/CLAIMS-TEMPLATE.md)"

See:

• ~/.claude/skills/proveryay/SKILL.md — hook behavior
• ~/.claude/skills/proveryay/CLAIMS-TEMPLATE.md — canonical format
• ~/.claude/skills/fleet-eval/SKILL.md — probe taxonomy