Skip to content
Back to search

geiger

A Geiger counter for AI agents — one read-only command that inventories every agent, MCP server, plugin, and AI extension on a machine

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/atomburstofficial--geiger/badge.svg)](/listings/atomburstofficial--geiger)

HTML

<a href="/listings/atomburstofficial--geiger"><img src="/api/listings/atomburstofficial--geiger/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 220 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # geiger
  2. [![CI](https://github.com/Atomburstofficial/geiger/actions/workflows/ci.yml/badge.svg)](https://github.com/Atomburstofficial/geiger/actions/workflows/ci.yml)
  3. [![npm](https://img.shields.io/npm/v/geiger-scan)](https://www.npmjs.com/package/geiger-scan)
  4. [![license: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
  5. ![node ≥18](https://img.shields.io/badge/node-%E2%89%A518-brightgreen)
  6. ![dependencies: 0](https://img.shields.io/badge/dependencies-0-success)
  7. **A Geiger counter for AI agents.**
  8. One read-only command that inventories every AI agent, harness, MCP server,
  9. plugin, and AI extension on a machine — and tells you, in plain language,
  10. what each one can touch.
  11. ```
  12. npx geiger-scan
  13. ```
  14. No install. No account. No telemetry. Reads configs and directories, writes
  15. nothing (unless you ask for `--json yourfile.json`).
  16. ![A geiger scan: findings grouped by ecosystem with exposure labels, a redacted credential, and plain-language fixes](docs/scan-terminal.svg)
  17. ## Why this exists
  18. In August 2026, an open-source agent harness went from zero to 200,000+
  19. GitHub stars in three weeks. Its plugin ecosystem passed 13,000 repositories
  20. in the same window. One-click desktop clients appeared the same day it
  21. launched. Instagram carousels now teach office workers to install all of it.
  22. Every one of those installs is a program that can execute commands, read
  23. files, and hold credentials — configured in dotfiles nobody looks at twice.
  24. Ask yourself the question this tool answers: **what is actually running on
  25. this machine, and what can it reach?** Most people cannot answer it. Now
  26. it's one command.
  27. ## What a scan looks like
  28. ```
  29. GEIGER · a Geiger counter for AI agents
  30. machine dev-laptop · 2026-09-06 12:24 UTC · read-only · no telemetry
  31. ──────────────────────────────────────────────────────────────
  32. 9 findings across 3 ecosystems · 7 can execute code · 1 credential in config files
  33. claude-code (6)
  34. Claude Code agent
  35. [EXECUTES] [BROAD-FILESYSTEM] [NETWORK]
  36. origin: registry · @anthropic-ai/claude-code
  37. magic (Claude Code · global) MCP server
  38. [EXECUTES] [HOLDS-SECRETS] [BROAD-FILESYSTEM]
  39. origin: registry · @21st-dev/magic@latest
  40. credential: "API_KEY" — opaque value under a credential-named key · ~/.claude.json
  41. note: wrapped by a policy agent (domainguard-agent.exe) — enforcement layer in front of the server
  42. hooks: UserPromptSubmit, PreToolUse hook
  43. [EXECUTES]
  44. note: hooks execute without a prompt each time their event fires
  45. ...
  46. ```
  47. Real output from a real machine (values redacted — see below).
  48. ## What it detects
  49. | Ecosystem | What geiger reads |
  50. |---|---|
  51. | **Claude Code** | global + per-project MCP servers, hooks, plugins, skills, subagents, `apiKeyHelper` |
  52. | **MCP hosts** | MCP servers configured in Claude Desktop, Cursor, Windsurf, VS Code (user + project), Cline, Roo Code, Continue, Zed |
  53. | **AI apps & IDEs** | Cursor, Windsurf, Zed, Claude Desktop, ChatGPT Desktop, Codex desktop app — reported by presence, so an installed client with zero MCP servers still shows up |
  54. | **Other agents** | Codex CLI, Gemini CLI, Kilo CLI, Grok Build, Aider, OpenCode, Qwen Code, DeepSeek Harness, Continue, GitHub Copilot CLI, Goose, JetBrains Junie, Open Interpreter, LM Studio, Ollama |
  55. | **Hooks** | Claude Code (settings), Cursor (`hooks.json`), Codex (`notify` in `config.toml`), Gemini CLI (settings) — the commands are listed, because hooks run with no prompt |
  56. | **Git hooks** | live hooks in `.git/hooks` (samples ignored), `core.hooksPath` redirects, and merge drivers — git runs these itself, and agent tooling installs itself here |
  57. | **Skill & subagent text** | what a skill actually says: its description, instruction text that tries to override the agent or hide work from you, commands with a hostile shape, and credential shapes — a name in a directory listing tells you nothing |
  58. | **Editor extensions** | AI extensions in VS Code / Insiders / Cursor |
  59. | **JetBrains IDEs** | AI Assistant / MCP settings presence per product (the settings live inside the IDE — geiger points you at the right screen) |
  60. | **Global CLIs** | agent packages in global npm roots (read directly — npm is never executed) |
  61. | **AI browsers** | Comet, Dia, ChatGPT Atlas — the browser is the agent, so its presence is a finding |
  62. | **Browser extensions** | AI extensions in Chrome / Edge / Brave / Vivaldi / Arc / Comet / Dia / Atlas / Firefox profiles, with their granted permissions |
  63. Every finding gets: what it is, where it came from (registry, store, git,
  64. local script, remote server — or **UNKNOWN-ORIGIN**), what it can do
  65. (**EXECUTES**, **HOLDS-SECRETS**, **BROAD-FILESYSTEM**, **BROAD-WEB**,
  66. **NETWORK**), and the evidence path so you can verify by hand.
  67. Geiger also recognizes **policy wrappers** (agents that put an enforcement
  68. layer in front of MCP servers) and reports both layers instead of hiding the
  69. real server behind the wrapper.
  70. ## The three promises
  71. 1. **Read-only.** The only write geiger ever performs is the `--json` file
  72. you explicitly name.
  73. 2. **No telemetry.** Nothing leaves your machine. There is no endpoint to
  74. send anything to. (This also means we have no idea how many people use
  75. this — a trade we're happy with.)
  76. 3. **Secrets by shape only.** When a credential-shaped value is found in a
  77. config, geiger reports the key name, the file, and what kind of secret it
  78. looks like — never any part of the value. A redaction pass runs on all
  79. output as defense-in-depth, and the test suite enforces it.
  80. ## Requirements
  81. **Node.js 18 or newer. That's it.** Node ships with `npm` and `npx`, and
  82. geiger has zero dependencies, so nothing else gets installed. No global
  83. install, no admin rights, no account.
  84. If Node isn't on the machine yet, it's one command with the package manager
  85. you already have — then open a new terminal so `npx` is on the path:
  86. ```
  87. winget install OpenJS.NodeJS.LTS # Windows
  88. brew install node # macOS (Homebrew)
  89. sudo apt install nodejs npm # Debian / Ubuntu
  90. ```
  91. Other platforms and version managers (nvm, fnm): https://nodejs.org/en/download
  92. ## Usage
  93. ```
  94. npx geiger-scan scan, print the report
  95. npx geiger-scan --html report.html self-contained HTML report with per-finding
  96. "what to do" remediation guidance
  97. npx geiger-scan --json out.json machine-readable findings (schemaVersion 1)
  98. npx geiger-scan --path D:\repo1 --path E:\repo2
  99. also scan these project directories for
  100. project-level agent and MCP configs
  101. npx geiger-scan --home C:\Users\other scan a different home root (another user
  102. profile, a mounted image)
  103. npx geiger-scan --strict exit 2 if anything can execute code or
  104. holds secrets
  105. npx geiger-scan --diff baseline.json compare against an earlier --json
  106. snapshot: what appeared, disappeared,
  107. or escalated since then
  108. ```
  109. **Drift alarm:** once you've reviewed a machine, save a baseline
  110. (`--json baseline.json`) and put `geiger-scan --strict --diff baseline.json`
  111. in cron or CI. It exits 2 only when something **new** can execute code or
  112. hold secrets — the standing, already-reviewed inventory stays quiet. Same
  113. mental model as a lockfile: accept what's there, alarm on change.
  114. Registry blocked, or want the unreleased `main`?
  115. `npx github:Atomburstofficial/geiger` runs straight from this repo (still
  116. needs Node — see [Requirements](#requirements)).
  117. ## What the reports look like
  118. Before running anything, see exactly what you'd get: a
  119. [sample HTML report](docs/sample-report.html) and a
  120. [sample JSON output](docs/sample-report.json) live in this repo, generated
  121. from the test fixture — synthetic data, generic paths, credentials shown by
  122. shape only (as always). The HTML report:
  123. ![The HTML report: summary tiles, exposure chips, credential shapes, and "What to do" remediation blocks](docs/sample-report.png)
  124. Every finding that warrants action carries plain-language remediation — as
  125. `fix:` lines in the terminal and "What to do" blocks in the HTML report.
  126. Fleet pattern (MSPs, IT): run with `--json` per machine on a schedule (an
  127. RMM task or login script writing `%COMPUTERNAME%.json` to a share), keep
  128. each machine's baseline, and let `--diff` report per-machine drift. The
  129. schema is versioned and stable. Geiger never phones home — the JSON files
  130. travel only where you put them.
  131. ## Limitations
  132. Stated up front, because a scanner you overtrust is worse than no scanner:
  133. - Geiger reads **known config locations**. Agents installed in nonstandard
  134. paths, other user accounts, containers, or WSL (from the Windows side)
  135. are not seen.
  136. - It reads **configuration, not runtime behavior**. It cannot tell you what
  137. a plugin actually did — only what its position allows.
  138. - It cannot judge whether a package is malicious — only where it came from
  139. and what it can reach. Origin ≠ trustworthiness.
  140. - Partially-parseable formats (TOML configs) and JSON configs that fail to
  141. parse are still reported: scanned for credential shapes, flagged with
  142. reduced confidence, and called out when they appear to declare servers or
  143. hooks — never silently skipped. Their individual servers and hooks are not
  144. listed until the file parses, so treat an "unparseable config" line as
  145. something to open by hand.
  146. - Hook bodies and skill text are matched by **shape**. A flagged line means
  147. "open this by hand", never "this is malicious", and the flags are
  148. heuristics: a hostile file written carefully enough will not trip them.
  149. Nothing geiger reads is ever executed.
  150. - Git hooks are read only for the directories you scan (the current directory,
  151. plus any `--path`), not for every repo on the machine.
  152. - AI browsers and desktop apps are reported by **presence** with
  153. medium confidence; their capabilities are inherent to the product, not
  154. read from a config. Profile paths for Dia and ChatGPT Atlas (macOS) are
  155. best-effort — a wrong guess finds nothing rather than something false.
  156. - The ecosystem this tool audits changes weekly. Detectors are data-driven
  157. and small on purpose — see [CONTRIBUTING.md](CONTRIBUTING.md) to add one.
  158. ## FAQ
  159. **Is this a security audit?** No. It's an inventory with honest exposure
  160. labels — the thing you need *before* any audit means anything.
  161. **Why should I trust a security company's free scanner?** Read it. It's a
  162. few hundred lines of dependency-free JavaScript, and what's published is
  163. what runs — releases are published from GitHub Actions with [npm
  164. provenance](https://docs.npmjs.com/generating-provenance-statements), so the
  165. npm page carries a signed link to the exact public commit each version was
  166. built from.
  167. **What do I do about what it finds?** Individually: remove what you don't
  168. recognize, rotate credentials that shouldn't be sitting in configs. At a
  169. company: that's policy enforcement, which is a different product —
  170. [DomainGuard](https://atomburst.io/domainguard) is how organizations put a
  171. policy layer in front of this surface. Geiger stays free and standalone
  172. either way.
  173. ## License
  174. [MIT](LICENSE) · built by
  175. [Atomburst](https://atomburst.io/geiger?utm_source=github&utm_medium=readme) ·
  176. zero runtime dependencies, no build step — the source you read is the code
  177. that runs.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.