adom-plan — Exhaustive Plan Review

End-to-end planning pipeline: expand a sketch into a detailed plan, then stress-test it via adversarial multi-model debate. Uses ultraplan when available (harness-level deep expansion), otherwise runs a built-in Opus expansion pass. Then launches three adversarial pairs (6 agents: Opus vs Sonnet per category) arguing architecture, implementation, and security from opposite dispositions. A 7th agent red-teams the consensus. Final output is a scored, vetted plan with validation strategies, E2E testing, risk register, and prioritized action items.

Total agents: up to 8 (1 expansion + 6 debate + 1 red-team consensus).

How it works

1. Phase 0 — Locate the plan (plan file, file path argument, or conversation context)
2. Phase 0.5 — Expand the plan via ultraplan (if available) or built-in Opus expansion agent
3. Phase 1 — Fetch platform context (wiki catalog, installed skills, recent commits, existing plans)
4. Phase 2 — Launch 6 adversarial debate agents in parallel (3 pairs):
   • Architecture: Opus advocate vs Sonnet challenger
   • Implementation: Sonnet optimist vs Opus pessimist
   • Security: Sonnet auditor vs Sonnet red-team attacker
5. Phase 3 — Red-team the consensus with one final Opus agent
6. Phase 4 — Scored synthesis with scorecard, contested decisions, risk register, E2E strategy, and action items
7. Phase 5 — Present summary to user

Key features

• Adversarial pairing — two agents argue each category from opposite dispositions, surfacing disagreements a single model would self-censor
• Multi-model diversity — Opus reasons conservatively and architecturally; Sonnet is pragmatic and detail-oriented
• Structured scoring — 1-5 scale across Feasibility, Maintainability, Platform Leverage, Simplicity; 2+ point deltas auto-flagged
• Platform awareness — fetches wiki catalog so agents reference real tools, not hypothetical ones
• The 30-Day Failure Scenario — forces concrete, actionable thinking about what will actually go wrong
• Consensus red-team — attacks the "safe" decisions everyone agrees on, because the most dangerous risks are the ones nobody flags

Invocation

/adom-plan                          # review the active plan file
/adom-plan path/to/plan.md          # review a specific file

Installation

Bundled with gallia. Available at ~/.claude/skills/adom-plan/ after install.mjs runs. The skill consists of two files:

• SKILL.md (212 lines) — workflow, phases, synthesis format, guidelines
• debate-prompts.md (413 lines) — all 8 agent prompt templates (read on demand during execution)

Output

The skill appends a Debate Review Summary to the plan file containing:

• Scorecard — per-section grades (A/B/C/F) with dimensional scores
• Consensus Decisions — verified safe items
• Contested Decisions — where agents disagreed, with recommendations
• Fragile Consensus — items the red-team flagged
• Risk Register — consolidated threat table with mitigations
• Validation & E2E Strategy — complete test plan
• The 30-Day Failure Scenario — the most likely production failure
• Action Items — numbered list tagged BLOCKING / RECOMMENDED / NICE-TO-HAVE