[Senate/feat] Comment classifier — LLM-tag each artifact comment {proposal,action,evidence,dispute,question,ack} done

Quest: Percolation Engine (f80549eb-38d). Foundation for all percolation: comments must be classified before they can trigger downstream actions. Acceptance: - On comment INSERT (trigger or post-write hook), call lightweight LLM classifier (haiku or local model) → set artifact_comments.comment_type ∈ {proposal, action, evidence, dispute, question, ack, other}. - Multi-label allowed (e.g., a post can be both 'evidence' and 'action'); store as JSON array if multi. - Confidence score persisted alongside; below-threshold posts left as 'other' until re-classified. - Idempotent: re-running classifier produces stable output for unchanged content. - Backfill: classify any historic comments once table is populated. - Spec: docs/planning/specs/percolation_comment_classifier.md. Why now: percolation rules need categorical signal on every post; without classifier, downstream emitters can't fire.

Git Commits (4)

{
  "completion_shas": [
    "91bffc1bb",
    "7236a9fc4",
    "297ff4ec0"
  ],
  "completion_shas_checked_at": ""
}

Sibling Tasks in Quest (Percolation Engine) ↗