Hyperplan (Adversarial Multi-Agent Planning)

What it does

hyperplan replaces consensus with intellectual combat. The orchestrator becomes the Lead of a five-member adversarial team spawned through team-mode:

• skeptic — attacks over-engineering, scope creep, premature abstraction
• validator — hunts edge cases, cross-module fragility, blast radius
• researcher — demands file:line or doc-URL evidence; kills vibes-based claims
• architect — exposes hidden coupling, leaky abstractions, technical debt
• creative — forces at least three lateral alternatives before any answer wins

Execution runs in 7 phases: restate the request → create the team → Round 1 independent findings → Round 2 ruthless cross-attack → Round 3 defend/refine/concede → Lead distills survivors into four buckets (hard constraints, decisions, risks & mitigations, open questions) → mandatory handoff to the plan agent for an executable plan, then team cleanup.

The key design choice: the Lead is forbidden from writing the plan. Only battle-tested insights go to a dedicated planner that adds sequencing, dependencies, and per-task success criteria.

Who it's for

• Engineers facing hard-to-reverse architecture calls or large refactors
• Anyone who wants blind spots and optimistic assumptions forcibly surfaced
• opencode users who already run team-mode (team_* tools) and multi-agent setups
• Overkill for small bug fixes — five agents across three rounds burns a lot of tokens.

Examples

1. Payment module refactor: "hyperplan a plan to split the payment domain" → skeptic deletes speculative abstraction layers, validator enumerates which existing tests break, and you get a staged migration plan.
2. New feature design review: "adversarial plan for realtime notifications" → creative forces a polling vs. webhook vs. SSE comparison, architect flags coupling, and only evidence-backed options survive into the plan.
3. Performance roadmap: "hpp: plan to cut API latency" → researcher demands actual profiling data, filtering out guess-based optimizations; unresolved disputes become explicit user-input gates in the final plan.

Install the skill I found on Claude Skill Mart.
Copy the .agents/skills/hyperplan folder from the GitHub repo code-yeongyu/oh-my-openagent into my ~/.claude/skills/hyperplan/.
When it's done, tell me in one line what this skill can do.