As a developer, I want to collect usage records via skill execution so that project teams can submit data with minimal effort

[kiyotis]

Situation

Usage records are collected from project teams to track nabledge-6 adoption, quality, and cost trends. A usage record format and storage rules have been defined in:

• Format and storage: .claude/rules/usage-records.md
• Storage location: docs/usage-records/

Records contain two types of data:

• Numeric metrics: scale (file count, line counts), elapsed time, tokens, cost
• Qualitative feedback: output quality, UX, performance impressions

Pain

Without automated collection, usage data requires users to manually record metrics and write free-form feedback — both are error-prone and unlikely to be done consistently.

Benefit

• Project teams can submit usage records with minimal effort
• Collected data enables analysis of cost, performance, and quality trends across versions and processing patterns

Success Criteria

Collection Mode (numeric data, runs at end of skill execution)

• Code analysis skill supports an opt-in collection mode
• Automatically captures: target file count, line counts (code/comment/blank), elapsed time (displayed), nabledge version, Nablarch version
• Asks user for processing pattern (e.g., バッチ処理, Web API) if not inferable
• Creates a usage record file in docs/usage-records/ following the naming convention in usage-records.md

Hearing Mode (qualitative feedback, run separately after reviewing output)

• Hearing mode is a separate skill invocation (e.g., /n6 record-interview)
• Loads the most recent usage record created by collection mode (or accepts a path parameter)
• Asks for the CLI summary line (Done (X tool uses · Xk tokens · Xm Xs)) to capture tokens, tool uses, and actual elapsed time; calculates cost automatically
• Asks qualitative questions one at a time per feedback category (出力品質, UX, パフォーマンス, その他)
• Questions are open-ended and suggestive without being leading
• AI formats answers into concise comments and appends to the usage record
• Questions are optimized per skill type (code analysis questions differ from future skills)
• Shows completed record file path at the end