coding-agent-tools 0.1.2

Session search and unified usage analytics for coding agents

CodingAgentTools

coding-agent-tools is a practical CLI bundle for local coding-agent workflows.

It currently provides:

• find-session: search and resume sessions across Claude Code, Codex, and OpenCode
• agent-usage: unified usage and estimated USD-equivalent cost across Claude Code, Codex, OpenClaw, OpenCode, and OpenWhispr

Why this repo exists

This consolidates earlier standalone tools into one package:

• find-session (now here)
• local allusage script (now agent-usage)

The goal is one install, one command surface, one place to improve.

Install

From PyPI (recommended):

uv tool install coding-agent-tools

Alternative:

pipx install coding-agent-tools
# or
pip install coding-agent-tools

From source:

uv tool install .

Quickstart

# Find sessions in current project across all supported agents
find-session "auth,refactor"

# Search all projects
find-session -g

# Usage report (daily default)
agent-usage

# Weekly report with per-model expansion
agent-usage weekly --breakdown

find-session

Unified session search with interactive selection, resume, and export/copy actions.

# Search current project
find-session "redis,bug"

# Search all projects
find-session -g

# Limit to one agent
find-session "checkpoint" --agents codex

Preview behavior

Session previews now include both ends of user intent:

• First: ...
• Last: ...

If only one meaningful user message is found, it displays as First/Last: ....

agent-usage

Token usage and estimated cost tables by period.

# Daily usage for all sources
agent-usage

# Weekly with per-model breakdown
agent-usage weekly --breakdown

# Restrict sources
agent-usage monthly opencode codex

Sources and aliases

• claude / cc
• codex / cx
• openclaw / oc / claw
• opencode / oe / code
• openwhispr / ow / whispr

Cost model

agent-usage estimates costs from token counts using live LiteLLM pricing metadata.

• This avoids misleading $0 totals for subscription-backed traffic (especially OpenCode).
• If a model is missing in the pricing registry, it falls back to source-observed cost when present.

Data sources

• Claude Code: ~/.claude/projects/*/*.jsonl
• Codex: ~/.codex/sessions/**/rollout-*.jsonl
• OpenCode: ~/.local/share/opencode/opencode.db
• OpenClaw: ~/.openclaw/agents/*/sessions/*.jsonl
• OpenWhispr: ~/Library/Application Support/open-whispr/transcriptions.db

Beautiful terminal demos (yes, via subprocess recording)

Yes: you can generate polished terminal renders by recording commands as subprocesses.

This repo includes ready-to-run VHS tapes in demo/:

# macOS
brew install vhs

# render animated gifs from deterministic command scripts
./scripts/render-demos.sh

How it works:

• vhs launches a shell subprocess
• runs scripted commands (sample queries and real command output)
• renders a styled terminal recording directly to GIF

Use these as README assets, release previews, or social clips.

find-session demo

agent-usage demo

Development

python3 -m pip install -e .
find-session --help
agent-usage --help

References

• ccusage (inspiration for usage/cost reporting style): https://github.com/ryoppippi/ccusage
• claude-code-tools (Cloud/Claude Code tools reference and prior consolidation work): https://github.com/charlesnchr/claude-code-tools
• LiteLLM model pricing registry: https://github.com/BerriAI/litellm/blob/main/model_prices_and_context_window.json

License

MIT