Changelog

What shipped, dated.

Releases across the platform, the console, and this site. Real dates from the record — if it's listed here, it's live.

2026-09-21 Open source

Geiger 0.4.0 — git hooks, and reading what a skill actually says

  • Git hooks and merge drivers. Git runs .git/hooks itself on commit, checkout, push and merge — no prompt, no approval — and agent tooling increasingly installs itself there. Geiger now reports live hooks with the programs each one invokes, core.hooksPath redirects, and merge-driver commands. Git's sample hooks never run, so they are not findings.
  • Skills are read, not just listed. A skill is instructions your agent loads and follows, so the text is the payload. Geiger now reports each skill's description and flags instruction text that tries to override the agent, hide work from you, or bypass approval — plus commands with a hostile shape and any credential shape, still by shape and never by value.
  • A flagged line means "open this by hand", never "this is malicious". The flags are heuristics, stated as such in the README, and nothing geiger reads is ever executed. Verified against a real agent-tooling install, not fixtures alone. Release notes.
2026-09-19 Open source

Geiger 0.3.1 — security fix: BOM-prefixed configs no longer vanish

  • Configs saved with a byte-order mark are read again. Notepad, older Visual Studio and PowerShell 5.1 add one by default, and it made JSON configs fail to parse: MCP servers and hooks in those files dropped out of the report, and a new hook could pass --diff --strict as "no drift". Upgrade if you use the drift alarm.
  • Unreadable configs are never skipped silently. Every reader now reports a file it cannot parse as an "unparseable config" finding, scans it for credential shapes, and says when it appears to declare servers or hooks.
  • Parse errors carry a line and column, nothing else. Before, a malformed config could copy up to about ten characters of a credential into the JSON and HTML reports through the error message.
  • Reported privately by an independent researcher, with an exact reproduction — thank you. Release notes.
2026-09-12 Open source

Geiger 0.3.0 — hosts by presence, AI browsers, hooks beyond Claude Code

  • Installed clients count even with nothing configured: Cursor, Windsurf, Zed, Claude Desktop, ChatGPT Desktop and the Codex desktop app are now findings by presence. Agentic IDEs are labeled EXECUTES; chat clients NETWORK only.
  • AI browsers: Comet, Dia and ChatGPT Atlas are reported when their profile exists — the browser is the agent. Their profiles, plus Vivaldi and Arc, join the AI-extension walk.
  • Hooks from Cursor, Codex and Gemini CLI alongside Claude Code's, with each command listed — hooks run with no prompt and no model in the loop.
  • Since launch: 0.2.0 added diff mode (--diff baseline.json --strict exits non-zero only on new hot findings — a drift alarm for cron or CI); 0.2.1 added Kilo CLI, Grok Build, Firefox and JetBrains coverage. Still zero dependencies, read-only, no telemetry. Release notes.
2026-09-06 Open source

Geiger — a Geiger counter for AI agents

  • Our first open-source release: Geiger is a free, MIT-licensed, read-only scanner. One command — npx geiger-scan — inventories every AI agent, harness, MCP server, plugin, and AI extension on a machine and reports what each one can touch.
  • Zero runtime dependencies, no build step, no telemetry; credential values are never printed, enforced by the test suite.
  • Source on GitHub, package on npm.
2026-08-12 Site

Light mode

  • A light theme, as an option: the toggle in the top bar switches the whole site to ink-on-paper and remembers your choice. Dark stays the default.
  • Product material — console screenshots, the signal-path diagram, the live-verdict panels — deliberately stays dark in both themes.
  • All colors re-checked for WCAG AA contrast in the light palette.
2026-08-11 Site

Product truth pass

  • Access Review & Evidence is now its own product page — GitHub posture & discovery, run from the Atomburst console. It was previously presented as an ITDR module; they're separate products.
  • The ITDR page now describes the live platform: continuous identity risk scoring, Microsoft Graph-powered signals, unified alert & incident response, and managed-device posture insights.
  • Desktop heroes across the site now lead with product proof — screenshots and live-signal panels — instead of empty space.
  • Clean product URLs: /itdr, /domainguard, /mdr, and /access-review — the legacy -light.html addresses redirect permanently.
2026-08-10 Brand Site Console

The "Amplified" identity

  • Full rebrand: new A-mark and burst logotype, Michroma + Saira type, near-black design system across every page — and the tagline: Security, amplified by intelligence.
  • Console restyled to match, including print-ready evidence reports.
  • Published a complete sample evidence report (synthetic specimen) plus real console screenshots across the site.
  • New MSP program page and this changelog.
2026-07-03 Site Console

Self-serve pricing, stated plainly

  • Access Review & Evidence at $15/month (3 scans included) went live on the pricing page — sign in with GitHub, no sales call.
  • Privacy policy and terms of service published; the security page states exactly what we practice, nothing more.
  • Console hardening: rate limiting on authentication and scan routes.
2026-06-16 Console

The review workflow, end to end

  • Live access-review scans: connect a GitHub org through our read-only App and pull real org signals — members, 2FA state, outside collaborators, app grants.
  • Decisions with names attached: KEEP / FLAG / REVOKE with justifications, formal objections, and a printable evidence report.
  • Persistent multi-tenant store (Postgres), admin tenant management with editable access rules, Stripe subscriptions with monthly scan quotas.
2026-06-15 Console

Console launch

  • Authenticated multi-tenant console at /app: GitHub-App sign-in, default-deny tenancy — an identity reaches a tenant only if its login or verified email domain is explicitly allowed.
2026-06-10 Site

Accessibility & trust pass

  • WCAG AA contrast tokens site-wide, reduced-motion support on every page, visible focus rings, self-hosted fonts (no third-party font origins in the CSP).
  • Live product-app log-in links, glossary (21 terms), per-page SEO and social cards.
2026-06-07 Site

atomburst.io is live

  • Launch on Railway behind Caddy with strict security headers (CSP, HSTS), sitemap + IndexNow, working contact form, mobile navigation.

See the newest release in person.

The fastest way to evaluate Atomburst is to run an access review on your own org — the whole flow above is self-serve.

$15/month · 3 scans included · Cancel anytime
Start now — sign in with GitHub See a sample report