All plugins

bumblebee-guard

Supply-chain guard: runs the local Bumblebee scanner bridge (read-only, no sudo), exact-matches every installed package, MCP server and editor/browser extension against curated catalogs of known-compromised packages, and adds OSV.dev advisory counts via the bridge's OSV layer. Tools: bumblebee_guard_scan / _status / _diff (baseline diff of findings). A rate-limited pre_llm_call hook injects a one-line supply-chain alert only when curated findings exist (never when clean, at most once per hour). /bumblebee slash command plus a bundled triage skill covering confidence classes, OSV alias deduplication and a documented false-positive pattern (scanner fixtures containing threat-intel strings). Disclosure — the plugin itself makes no network calls and holds no credentials; the wrapped bridge's scan queries OSV.dev and an opt-in, off-by-default catalog refresh fetches threat-intel JSON from the upstream perplexityai/bumblebee repository; scans run daily_scan.sh as the current user and read package metadata in $HOME (never .env or token files, never executes package managers); a full scan takes up to ~3 minutes and honors a configurable timeout; the hook fails open and never blocks turns, so cron and unattended runs are unaffected.

kahnanXToolsv0.1.1Updated 4 hours ago

About

Supply-chain guard: runs the local Bumblebee scanner bridge (read-only, no sudo), exact-matches every installed package, MCP server and editor/browser extension against curated catalogs of known-compromised packages, and adds OSV.dev advisory counts via the bridge's OSV layer. Tools: bumblebee_guard_scan / _status / _diff (baseline diff of findings). A rate-limited pre_llm_call hook injects a one-line supply-chain alert only when curated findings exist (never when clean, at most once per hour). /bumblebee slash command plus a bundled triage skill covering confidence classes, OSV alias deduplication and a documented false-positive pattern (scanner fixtures containing threat-intel strings). Disclosure — the plugin itself makes no network calls and holds no credentials; the wrapped bridge's scan queries OSV.dev and an opt-in, off-by-default catalog refresh fetches threat-intel JSON from the upstream perplexityai/bumblebee repository; scans run daily_scan.sh as the current user and read package metadata in $HOME (never .env or token files, never executes package managers); a full scan takes up to ~3 minutes and honors a configurable timeout; the hook fails open and never blocks turns, so cron and unattended runs are unaffected.

bumblebee-guard

Supply-chain guard plugin for Hermes Agent.
Wraps the local Bumblebee scanner
bridge with Hermes tools, a rate-limited context-injection hook, a slash
command and a bundled triage skill.

Bumblebee inventories installed packages (npm, PyPI, Go, RubyGems, Composer),
MCP server configs and editor/browser extensions, and exact-matches every
(ecosystem, name, version) against curated catalogs of known-compromised
packages (Shai-Hulud variants, gemstuffer, node-ipc stealer, …). An optional
OSV layer matches installed versions against OSV.dev advisories.

What it does

  • bumblebee_guard_scan — run a full read-only scan (no sudo, no package
    managers executed). 1–4 minutes.
  • bumblebee_guard_status — latest results without re-scanning: curated
    findings, new-since-last, OSV count, packages inventoried, catalogs loaded.
  • bumblebee_guard_diff — diff curated findings against a stored baseline
    (show) or set the current state as baseline (set_baseline).
  • pre_llm_call hook — when the latest scan has curated findings, injects
    a one-line alert into the agent's context. Rate-limited to once per hour.
    Never injects when clean — zero context cost in the normal case.
  • /bumblebee — slash command: status (default) | scan |
    diff [show|set_baseline] | help.
  • Bundled triage skill — how to interpret findings: curated matches are
    high-confidence; OSV advisories need dedup and applicability checks;
    documented false-positive pattern (scanner fixtures containing
    threat-intel strings).

Requirements

  • Hermes Agent >= 0.21.6
  • The Bumblebee bridge deployed at ~/.local/share/hermes-bumblebee-bridge/ —
    installed from
    Deconstruct2021/hermes-bumblebee-bridge
    (maintained by @Deconstruct2021; this plugin is tested against commit
    d89dd95, bridge v0.1.0). Without it, tools
    report bridge not deployed cleanly — they never fail silently.
  • Python stdlib only. No credentials, no requires_env.

Configuration (plugins.entries.bumblebee-guard.settings)

| Key | Default | Meaning |
|---|---|---|
| bridge_dir | ~/.local/share/hermes-bumblebee-bridge | Bridge install dir |
| scan_timeout_s | 240 | Scan timeout |
| inject_status | true | Enable the context injection hook |
| osv_enabled | true | Include OSV advisory counts in reports |

Install

hermes plugins install <owner>/bumblebee-guard
hermes plugins enable bumblebee-guard

Disclosure (catalog rule 13)

  • The plugin itself makes no network calls and holds no credentials.
  • The wrapped bridge's scan makes its own calls: the optional OSV layer
    queries OSV.dev; an optional, off-by-default catalog refresh fetches
    threat-intel JSON from the upstream Bumblebee repository.
  • Scans run daily_scan.sh as the current user (no sudo) and read package
    metadata in $HOME; they never read .env/token files and never execute
    package managers.
  • Scan durations up to ~3 minutes; the tool honors scan_timeout_s and
    reports timeouts instead of hanging the agent. Cron/unattended runs are
    unaffected: the hook never blocks turns.

Tests

python3 -m pytest tests/ -q

License

MIT

README from the reviewed commit 8cdf7e0; it updates when the author re-pins.