Pi 编码智能体扩展 pi-token-usage：分析 token 用量与成本并按需清理旧会话

A Pi coding agent extension that

• analyzes token usage and cost across your session files
• prunes old sessions files on demand

Installation

# NPM (recommended)
pi install npm:@alexanderfortin/pi-token-usage

# Git
pi install git:github.com/shaftoe/pi-token-usage

# Or quick-test without installing
pi -e ./pi-token-usage/src/index.ts

Usage

Report Command

/token-report                                  Report all sessions (table in TUI)
/token-report 7                                Last 7 days
/token-report /path/to/dir                     Specific directory or file
/token-report --format csv                     CSV to stdout
/token-report --format csv --save report.csv   Save CSV to file
/token-report 7 --format json                  JSON to stdout
/token-report --format md                      Markdown to stdout
/token-report --format markdown --save out.md  Save Markdown to file
/token-report --format=csv --save=out.csv      Equals-sign syntax also works

Prune Command

/token-prune 30                    Delete sessions older than 30 days
/token-prune 30 --dry-run          Show what would be deleted without actually deleting
/token-prune 60 --path /custom/dir Prune a specific directory
/token-prune 100 --force           Skip confirmation prompt for large deletions

Report Arguments

Prune Arguments

Prune Behavior

The prune command performs two operations:

1. Old session deletion: Removes .jsonl session files whose modification time is older than the specified number of days
2. Empty file cleanup: Removes any .jsonl files that are empty (0 bytes), regardless of age

Safety features:

• Dry-run mode lets you preview what will be deleted
• If more than 100 files would be deleted, you must use --force to confirm (unless in dry-run mode)
• Only .jsonl files are affected; other files are ignored

Output Formats

Table (default) — interactive TUI overlay with styled columns.

CSV — machine-readable spreadsheet format. Includes metadata as # comment lines.

Model,Provider,Turns,Input Tokens,Output Tokens,...
claude-sonnet-4-20250514,anthropic,10,50000,12000,...
TOTAL,,15,70000,20000,...

JSON — structured output with meta, models, and totals sections. Pipe to jq for queries.

Markdown — pipe-delimited table for notes, PRs, or documentation.

Development

cd pi-token-usage
pnpm install         # Install dependencies
pnpm run check       # Type-check with tsc
pnpm run lint        # Lint with eslint
pnpm run format:check  # Verify formatting
pnpm run format      # Auto-format with Prettier
pnpm test            # Run tests
pnpm run test:coverage  # Run tests and show coverage stats

How It Works

Report Generation

1. Scans ~/.pi/agent/sessions (or $PI_CODING_AGENT_DIR/sessions when set, or a given path) for .jsonl session files
2. Parses each file, extracting assistant messages with usage data
3. Aggregates token counts and costs per model+provider
4. Renders in the requested format

Session Pruning

1. Scans the target directory (recursively) for .jsonl session files
2. Checks each file's modification time and size
3. Deletes files older than the specified days (or empty files)
4. Reports total files deleted, space freed, and any errors encountered

Releasing

This project uses automated publishing to NPM via GitHub Actions. The workflow will:

• Run all CI checks
• Build the package
• Publish to NPM with provenance (signed) via trusted publishing

License

See LICENSE