v0.2.1 — coverage expansion

Four ecosystems added, prompted by launch-day feedback:

• Kilo CLI — agent detection plus full MCP-server parsing from
  ~/.config/kilo/kilo.jsonc (JSONC comments handled) and project-level
  .kilo/kilo.jsonc / kilo.jsonc.
• Grok Build (xAI) — ~/.grok home with config.toml scanned for
  secret shapes and MCP declarations.
• Firefox — AI extensions per profile from extensions.json, with
  granted permissions mapped the same way as Chromium browsers
  (broad origins → BROAD-WEB, nativeMessaging → EXECUTES).
• JetBrains — AI Assistant / MCP settings detection per IDE product.
  JetBrains stores MCP config inside the IDE settings store rather than a
  documented file, so geiger reports that it exists, where, and which
  settings screen to review (medium confidence, honestly labeled). The
  Junie agent's ~/.junie home is detected alongside the other agent CLIs.

No behavior changes to existing detectors. Read-only, no telemetry, secrets
by shape only — unchanged, as always.

v0.2.0 — diff mode: the drift alarm

New

• --diff baseline.json — compare a fresh scan against an earlier
  --json snapshot and see what appeared, disappeared, or
  escalated since then. A server that gained a credential shows as
  CHANGED, not as a remove/add pair.
• Drift-gated --strict — with a baseline, --strict exits 2 only on
  new findings that can execute code or hold secrets. The reviewed
  inventory stays quiet. Same mental model as a lockfile: accept what's
  there, alarm on change. Built for cron and CI.
• The drift section renders in the terminal report, the --html report,
  and is included in --json output for fleet tooling.
• Sample reports in the repo — sample-report.html
  and sample-report.json, generated from the synthetic
  test fixture, so you can see exactly what to expect before running
  anything.

Unchanged, on purpose

Read-only. No telemetry. Secrets by shape only, enforced by test. Zero
runtime dependencies. Baselines travel only where you put them.

v0.1.1 — npm page refresh, provenance publishing

README refresh on the npm page (badges, terminal render), homepage now points at https://atomburst.io/geiger, and releases are published from GitHub Actions with npm provenance from this version on.

v0.1.0 — first public release

v0.1.0 — first public release

A Geiger counter for AI agents. One read-only command that inventories
every AI agent, harness, MCP server, plugin, and AI extension on a machine —
and tells you, in plain language, what each one can touch.

npx geiger-scan

What's in 0.1.0

• Six detector families: Claude Code (MCP servers global + per-project,
  hooks, plugins, skills, subagents, apiKeyHelper), MCP hosts (Claude
  Desktop, Cursor, Windsurf, VS Code, Cline, Roo Code, Continue, Zed), other
  agents (Codex CLI, Gemini CLI, Aider, Goose, Copilot CLI, DeepSeek
  Harness, Open Interpreter, LM Studio, Ollama, and more), VS Code AI
  extensions, global npm agent CLIs (read directly — npm is never executed),
  and browser AI extensions with their manifest permissions.
• Exposure labels, not scores: EXECUTES · HOLDS-SECRETS ·
  BROAD-FILESYSTEM · BROAD-WEB · NETWORK · UNKNOWN-ORIGIN.
• Secrets by shape only — key name, file, and shape; never the value.
  A redaction pass runs on all output and the test suite enforces it.
• Policy-wrapper awareness: wrapped MCP servers report both the
  enforcement layer and the effective server behind it.
• Reports: colored terminal report, self-contained --html report with
  per-finding "What to do" remediation, versioned --json (schemaVersion 1)
  for fleet diffing.
• Scoping flags: --path (repeatable) for extra project directories,
  --home for another user profile or mounted image, --strict for CI
  (exit 2 on EXECUTES/HOLDS-SECRETS).
• Zero runtime dependencies, no build step, no telemetry. CI on
  Linux/macOS/Windows × Node 18/20/22.

Known limitations

Stated in full in the README:
known config locations only; configuration, not runtime behavior; origin ≠
trustworthiness.