honne — Persona Synthesis

When invoked, execute Step 1 through Step 5 in order immediately. Do not summarize the skill or ask what the user wants — invocation itself is the request. Start by asking the Step 1 question.

Step 1: Locale HITL

Invoke AskUserQuestion tool:

• question: "Locale?"
• options: [{"label":"ko","description":"한국어"},{"label":"en","description":"English"},{"label":"jp","description":"日本語"}]

Set LOCALE from the reply. Do not use plain-text Q&A — arrow-key selection only.

Step 2: Load and Validate

Check that persona.json exists:

bash "${CLAUDE_PLUGIN_ROOT}/scripts/honne" persona check --persona .honne/persona.json

• Exit 66 → tell user: ".honne/persona.json not found. Please run /honne:whoami first to generate your persona." Stop.
• Exit 0 → proceed.

Check staleness:

python3 -c "import json,datetime; d=json.load(open('.honne/persona.json')); ts=datetime.datetime.fromisoformat(d.get('generated_at','2000-01-01T00:00:00Z').replace('Z','+00:00')); print((datetime.datetime.now(datetime.timezone.utc)-ts).days)"

Capture stdout as STALE_DAYS.

If STALE_DAYS exceeds 7 (or the value of HONNE_PERSONA_STALE_DAYS if set): warn "persona.json last updated {STALE_DAYS} days ago — consider re-running /honne:whoami." Then continue.

Step 3: Build Conflict Payload

Read .honne/persona.json in your mental context. Construct CONFLICT_PAYLOAD as a JSON object without writing to any file or using heredocs:

CONFLICT_PAYLOAD = {
  "locale": "<LOCALE>",
  "antipattern": {
    "claim": "<axes.antipattern.claim>",
    "explanation": "<axes.antipattern.explanation>",
    "evidence_strength": <axes.antipattern.evidence_strength>
  }  — or null if antipattern axis is absent or claim is null,
  "signature": {
    "claim": "<axes.signature.claim>",
    "explanation": "<axes.signature.explanation>",
    "evidence_strength": <axes.signature.evidence_strength>
  }  — or null if signature axis is absent or claim is null,
  "supporting_axes": {
    "<axis>": {"claim": "...", "explanation": "...", "evidence_strength": <val>}
    for each of the 5 remaining axes: lexicon, reaction, workflow, obsession, ritual
  }
}

Do NOT use python3 << 'EOF' or any heredoc. Assemble the payload in your mental context from the persona.json data you just read.

Step 4: LLM Synthesis

(a) Read synthesis prompt:

Read "${CLAUDE_PLUGIN_ROOT}/skills/persona/templates/persona_synthesis_prompt.${LOCALE}.md"

(b) Apply the synthesis prompt system instructions to yourself, with CONFLICT_PAYLOAD as the user input. Produce a STRICT JSON response:

{
  "conflict_present": true,
  "persona_antipattern": {
    "name": "...",
    "oneliner": "...",
    "system_prompt": "..."
  },
  "persona_signature": {
    "name": "...",
    "oneliner": "...",
    "system_prompt": "..."
  },
  "judge_system_prompt": "..."
}

Branch rules (enforced by synthesis prompt template):

• conflict_present = true: both axes present → generate two separate personas (antipattern and signature) + judge. All three fields REQUIRED.
• conflict_present = false, one axis null: set absent persona to null. judge_system_prompt is null.
• conflict_present = false, both axes null: all persona fields are null.

Constraints: each system_prompt ≤ 1000 tokens. judge_system_prompt ≤ 500 tokens. name ≤ 12 chars. oneliner ≤ 25 words.

(c) Save result: Write the JSON to {PWD}/.honne/cache/persona-synthesis.json. If JSON parse fails or response is empty, skip saving and output raw text with a warning.

Step 5: Render Personas

bash "${CLAUDE_PLUGIN_ROOT}/scripts/honne" render personas --persona .honne/persona.json --synthesis .honne/cache/persona-synthesis.json --locale "$LOCALE" --out-dir .honne/personas

Non-zero exit → output the raw synthesis fields with a warning that file render failed.

Output to user — choose the template matching the synthesis JSON state you saved in Step 4(c). Use LOCALE (ko/en/jp).

Case A — `conflict_present=true` (both personas + judge):

• ko: 두 인격이 생성되었습니다:\n- .honne/personas/antipattern.md — {persona_antipattern.name}\n- .honne/personas/signature.md — {persona_signature.name}\n- .honne/personas/judge.md — 심판자\n\n두 인격을 붙이려면 /honne:crush <주제>를 실행하세요.
• en: Two personas generated:\n- .honne/personas/antipattern.md — {persona_antipattern.name}\n- .honne/personas/signature.md — {persona_signature.name}\n- .honne/personas/judge.md — judge\n\nRun /honne:crush <topic> to stage a live debate.
• jp: ふたつのペルソナが生成されました:\n- .honne/personas/antipattern.md — {persona_antipattern.name}\n- .honne/personas/signature.md — {persona_signature.name}\n- .honne/personas/judge.md — 審判者\n\nライブ討論を行うには /honne:crush <テーマ> を実行してください。

Case B — `conflict_present=false`, exactly one persona non-null:

• ko: 하나의 인격만 생성되었습니다 (반대 축이 감지되지 않음):\n- .honne/personas/{slot}.md — {persona.name}\n\n/honne:crush 토론은 두 축이 모두 필요합니다. 더 많은 세션을 수집한 뒤 /honne:whoami를 재실행하세요.
• en: Only one persona generated (opposite axis not detected):\n- .honne/personas/{slot}.md — {persona.name}\n\n/honne:crush debate requires both axes. Collect more sessions and re-run /honne:whoami.
• jp: ペルソナが1つだけ生成されました (反対軸が検出されませんでした):\n- .honne/personas/{slot}.md — {persona.name}\n\n/honne:crush ディベートには両軸が必要です。さらにセッションを収集し /honne:whoami を再実行してください。

Where {slot} is antipattern or signature depending on which is non-null.

Case C — `conflict_present=false`, both personas null:

• ko: 인격이 생성되지 않았습니다 (두 축 모두 감지되지 않음). 세션을 더 수집한 뒤 /honne:whoami를 재실행하세요.
• en: No personas generated (neither axis detected). Collect more sessions and re-run /honne:whoami.
• jp: ペルソナは生成されませんでした (両軸とも検出されず)。セッションを追加収集し /honne:whoami を再実行してください。

IMPORTANT: This skill only generates files. Do not claim the persona is running, applied, or in use. Personas are independent artifacts — users invoke /honne:crush for live debate.