| name | team-shinchan:fierce-debate |
| description | Deterministic adversarial debate for high-stakes or irreversible decisions — mandatory refutation plus a scored judge panel. Opt-in main-loop Workflow tier. |
| user-invocable | false |
EXECUTE IMMEDIATELY
Fierce debate is a main-loop Workflow that GUARANTEES adversarial stress-testing — a non-skippable refutation round and schema-validated output that rejects empty dissent. The Midori/Task path (team-shinchan:debate) can only request this at the prompt level; this tier enforces it. Use ONLY for irreversible / high-stakes decisions (schema migration, public API contract, security boundary, data-loss risk). The cheap consensus default stays team-shinchan:debate.
Step 0: Validate + opt-in
- If args lack a topic or fewer than 2 competing options: ask the user for the decision and the options, then STOP.
- This skill calls the Workflow tool, which is main-loop only. Invoking
/team-shinchan:fierce-debate IS the explicit opt-in. Never delegate this skill to a subagent (Midori/Shinnosuke) — workflow() throws inside a Task child.
Step 1: Check past decisions
Read .shinchan-docs/debate-decisions.md. If an Active decision matches the topic, ask the user to reuse it or re-debate (carry the prior decision as context). This is the SAME ledger Midori uses.
Step 2: Run the fierce-debate Workflow
Give each option a persona — a ROLE/EXPERTISE descriptor drawn from the matching team-shinchan agent. (The Workflow runtime cannot load plugin subagents, so the persona is injected into the prompt, not passed as agentType.) Map each option to a lens — Backend/API → Buriburi; Frontend/UI → Aichan; DevOps/Infra → Masao; Architecture/general → Hiroshi; hidden requirements → Misae — then resolve that agent's persona string from its definition (DRY) with node ${CLAUDE_PLUGIN_ROOT}/src/workflow-personas.js <agent> and pass it as the option's persona. For each option's lens agent, ALSO resolve its learning block with node ${CLAUDE_PLUGIN_ROOT}/src/workflow-personas.js --learnings <agent> and inject it as that option's learnings field — distinct from persona, additive (FR-5/AC-8). Then call the Workflow tool with the shipped script:
Workflow({
scriptPath: "${CLAUDE_PLUGIN_ROOT}/skills/fierce-debate/fierce-debate.workflow.js",
args: {
topic: "<the decision>",
category: "<architecture|security|performance|tech-selection>",
options: [
// persona = the string printed by `workflow-personas.js <agent>` for the option's lens
{ label: "<Option A>", persona: "<output of workflow-personas.js hiroshi>", learnings: "<output of workflow-personas.js --learnings hiroshi>" },
{ label: "<Option B>", persona: "<output of workflow-personas.js misae>", learnings: "<output of workflow-personas.js --learnings misae>" }
]
}
})
The script runs: advocates (max case per option) → mandatory cross-refutation (every advocate attacks every other option, never skipped) → an Action-Kamen-prompted rubric judge. It returns { verdict: { scores, winner, rationale, dissenting_views }, advocates, refutations }.
Step 3: Record to the unified ledger
Append a [DECISION-NNN] entry to .shinchan-docs/debate-decisions.md using the shared format in ${CLAUDE_PLUGIN_ROOT}/agents/_shared/debate-decisions.md — with Tier: fierce plus Refutations, Judge Scores, Winner, and a non-empty Dissent. Write the transcript to .shinchan-docs/debates/DECISION-{NNN}.json. If .shinchan-docs/ontology/ontology.json exists, suggest a Decision entity (same as Midori). Never fork the ledger or schema.
Step 4: Confirm
Present the winner, the score table, and the surviving dissent. Ask the user to accept. Never finalize without confirmation — on disagreement, re-run or revise reflecting their concern.
