Official Claude-Mem demonstration: a coding session beside a browser showing persistent observations.
プロジェクトのスクリーンショットこのスキルについて
Analyze local agent transcripts into an HTML cost report, JSON, CSV line items, and evidence with explicit estimate labels.
おすすめする理由
A reusable workflow from the Claude-Mem project. It is listed separately to explain this task; shared runtime and helper dependencies remain part of the upstream installation. The inherited gallery demonstrates the memory system, not this task’s specific output.
役立つこと
- Understand a session or period’s token usage and review work categories and possible rework.
How the workflow runs
Step 1
Choose a project, session, or date range; the default is seven complete Pacific-time days excluding today.
Step 2
Measure transcript tokens, apply saved list-price evidence, and distinguish measured, estimated, and unavailable spending.
Step 3
Review draft work labels against evidence and render the report plus machine-readable files.
Expected result
HTML report, report.json, line-items.csv, evidence.json; PDF depends on the documented rendering path.
Try a concrete request
Example request
After installing the required integration, ask your agent:
Prepare last week’s agent usage report and clearly label list-price estimates versus measured provider charges.
使う前に
- Bundled Python scripts, Python 3.9+ with timezone data, local Claude Code or Codex transcripts, and network access for list prices. Memory review needs Claude-Mem.
- Use the original upstream files and follow the documented host integration. Model or third-party service usage can carry separate costs.
対応エージェント
資料・参照元
元の資料は以下のリンクから確認できます。実践のヒントは SkillHey による編集上の提案です。