Skip to content

As a developer, I want to collect usage records via skill execution so that project teams can submit data with minimal effort #181

@kiyotis

Description

@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

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions