language-polishinglisted

# Skill-7: Exclusive Language Polishing & Final Generation Input: `agent_model` (e.g. `RIDE`) + `use_case` (e.g. `Salary_Negotiation`) + `filled_data` (JSON, validated by Skill-3) + optional `user_revision_request` (e.g. "Make the tone less aggressive") Output: strict JSON — `{ "final_text": "...", "word_count": N, "hallucination_gate": "PASS" | "WARN" | null, "gap_note": "..." | null }` ## Procedure 1. **Load the model's generation prompt** — Read the `## Model reference` section in your agent instructions and extract the **Generation prompt** (or Read `config/model-reference.md` for the same text). Done when the model's synthesis prompt is in context; if missing → stop, report `{"action": "error", "reason": "Model <agent_model> has no generation prompt"}`. 1.5. **Validate the input before generating (Skill-12 input gate)** — Run the hallucination gate on the filled card data (the script lives in the sibling `hallucination-check` skill folder — from this skill's directory, `../hallucination-check/scripts/hallucination-gate.py`): `python ../hallucination-check/scripts/hallucination-gate.py --mode input --data "<the filled card values>"` (or `--data-file <path>` — the saved cards file; on Windows prefer the stdin pipe with `--data` first for non-ASCII safety). - `PASS` → proceed to Step 2. - `WARN` → proceed to Step 2 but confirm the flagged items (placeholders must become real values) with the user at the earliest natural point. - `BLOCK` → do NOT gene