Third-Party Tools

Cost Tracking

The tools in this card are adjacent layers: they observe, configure, or coordinate sessions. They do not replace the runtime that owns the agent loop. Use the Agent Harness Map to choose a runtime or orchestrator.

ccusage: the community reference tool for tracking API spending. Parses local JSONL files, zero data sent externally.

npx ccusage          # Today's summary
ccusage daily        # Daily report
ccusage blocks --live # Real-time monitoring (5h windows)
ccusage --breakdown  # Breakdown by model

ccburn: Python alternative with burn-rate charts in the terminal. Useful for visualizing consumption relative to billing limits, within a Python ecosystem.

pip install ccburn
ccburn

Dashboard & Monitoring

ccboard: TUI and web interface for monitoring sessions, costs, and statistics in real time.

cargo install ccboard
ccboard          # TUI interface
ccboard --web    # Web UI on localhost:3000

Social Sharing & Leaderboard

Straude: social dashboard to track and share consumption stats. Pushes daily metrics (cost, tokens, models) to a public leaderboard. Note: sends machine hostname and a device UUID to Straude servers. Use --dry-run before the first push.

npx straude@latest   # Auth + sync in one command
straude push --dry-run  # Preview without sending
straude status       # Streak, global rank, totals

viberank: leaderboard covering more tools (Claude Code, Codex, Gemini CLI, OpenCode, and more). Sends aggregate totals only, no hostname or device UUID. Publishes its dataset via a free JSON API under CC BY 4.0.

npx viberank-cli      # Submit ccusage data

Token Optimization

RTK (Rust Token Killer): CLI proxy that filters outputs before they enter the context. 92% reduction on git log, 90% on tests. See card T20 for details.

brew install rtk-ai/tap/rtk
rtk init             # Setup with automatic hook

Parallel Orchestration

Conductor: macOS application for orchestrating multiple Claude Code instances in parallel via git worktrees. Integrated interface: diff viewer, CI/GitHub Actions, Linear, PR workflow.

Install from conductor.build (macOS only as of March 2026).

Typical flow: create one workspace per feature from a GitHub or Linear issue, let the agents run, use “Next Workspace” to handle only those waiting for input.

Recommendation Table

Known Gaps

No visual editor for .claude/skills/ skill files. No unified dashboard combining config, sessions, costs, and MCP. A dedicated TUI exists for MCP servers (claude-code-config, pip install claude-code-config), but it stays limited to the ~/.claude.json scope: no settings.json, hooks, or commands.

These projects evolve independently from Claude Code. Check the repository, permissions, and targeted release before installation; the glossary distinguishes a runtime, orchestrator, protocol, and adjacent tool.