v0.1.0 — Initial public release

v0.1.0 — Initial public release

GhostSpend finds the AI spend you didn't know you had. This first release ships both install paths, the full setup → audit workflow, and the core baseline-comparison feature that makes it more than a token counter.

What's included

Core feature: known-tools baseline

• One-time setup (/gs-setup or ./scripts/setup.sh) records which AI CLI tools you actually, knowingly use.
• Every audit compares live spend against that baseline and flags anything unexpected — front and center, not buried in a combined total.

What it checks

• Global hooks (~/.claude/settings.json) firing on every tool call
• MCP server connectivity (claude mcp list) — failed or duplicated servers
• Plugin build integrity — cached plugins shipped as TypeScript source that never got built
• Project-level config drift across repos (.claude/settings.json, .mcp.json)
• Cross-provider spend via ccusage (Claude Code, Codex CLI, Gemini CLI, OpenCode, and more)
• Unexpected-tool flagging against your baseline
• Orphaned AI CLI processes left running after a crashed connection

Two ways to install

Option A — Claude Code plugin (recommended)

• Manual copy into ~/.claude/plugins/ghostspend
• Terminal marketplace commands: /plugin marketplace add danmackenz/ghostspend + /plugin install ghostspend
• Claude Desktop GUI: Settings → Plugins → Add → Add marketplace → Add from a repository

Option B — Standalone scripts

• No Claude Code required
• bash 3.2-compatible by design — runs on stock macOS with no Homebrew bash upgrade needed
• ./scripts/setup.sh then ./scripts/ghostspend.sh

Setup → audit handoff

• Setup now ends by asking directly: "GhostSpend setup complete. Ready to run an audit? (yes/no)"
• A yes answer runs the audit immediately in the background — no need to separately type /gs-audit.

Known limitations

• On-demand audit only — Claude Code hooks don't currently receive live token/cost data, so this isn't real-time monitoring.
• Doesn't auto-fix anything; every write action or install requires explicit confirmation.
• Codex CLI spend figures are ccusage estimates from token counts against third-party pricing data (LiteLLM), not an OpenAI-confirmed bill.
• Windows isn't natively supported yet — run via WSL. Native PowerShell support is tracked in ROADMAP.md.

Documentation

• Full install, usage, and troubleshooting guide in README.md
• Config schema reference in docs/config.md
• Worked example in examples/sample-audit-output.md
• Contribution guide in CONTRIBUTING.md

Contributing

Particularly useful contributions for the next release: a Windows-native PowerShell port, provider-specific drill-downs, root-cause tracing for background tool invocations, and historical spend trend tracking. See ROADMAP.md for the full list and what's explicitly out of scope.`