Official Claude-Mem demonstration: a coding session beside a browser showing persistent observations.
Project screenshotsAbout this Skill
Analyze local agent transcripts into an HTML cost report, JSON, CSV line items, and evidence with explicit estimate labels.
Why we picked it
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.
What it helps with
- 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.
Before you use it
- 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.
Supported agents
Sources & references
Original sources are linked below. Practical starting points are editorial suggestions from SkillHey.
