What problem does it solve?
When an operator corrects the agent's Japanese phrasing mid-conversation, that feedback is normally lost. This Skill captures those corrections as structured, pending writing-rule proposals so they can be reviewed and promoted into a persistent linting dictionary later.
Core Features & Use Cases
- Correction Detection: Triggers only when the operator rewrites a sentence, explicitly calls out a style problem, or asks to register a phrasing as a rule.
- Schema-Validated Proposals: Writes one JSONL record per correction with id, RE2 pattern, suggested replacement, evidence, and status pending, validated against the writing-rule-proposal.v1 schema.
- Strict Safety Boundary: It is the only writer of proposals.jsonl, never edits rules.jsonl, and never auto-promotes rules; promotion happens only via the human-run writing-rule-approve.sh script.
- Use Case: An operator says "その言い方はNG、こう書いて" during a session. The Skill extracts the before/after phrasing, builds a regex pattern, appends a pending proposal, and tells the operator the approval command.
Quick Start
Tell the agent to turn the Japanese phrasing correction you just made into a pending writing-lint proposal for later review.
