Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add Szotasz/marveen --skill bumblebee-hygiene-scangit clone --depth 1 https://github.com/Szotasz/marveenWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/szotasz/marveen/bumblebee-hygiene-scan)<a href="https://agentmods.dev/skills/szotasz/marveen/bumblebee-hygiene-scan"><img src="https://agentmods.dev/badge/skills/szotasz/marveen/bumblebee-hygiene-scan/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/szotasz/marveen/bumblebee-hygiene-scan"><img src="https://agentmods.dev/badge/skills/szotasz/marveen/bumblebee-hygiene-scan.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00056 | $0.02078 |
| Opus 5 | $0.00028 | $0.01039 |
| Sonnet 5 | $0.00011 | $0.00416 |
| Haiku 4.5 | $0.00006 | $0.00208 |
Grade C, and why
bumblebee-hygiene-scan scanned grade C with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
rm -rf /tmp/bb-synth How it starts
The opening of the file, as written. The whole thing — 124 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Bumblebee weekly supply-chain scan
When / purpose
Monday 09:00. The fleet uses many third-party MCP servers, auto-installed CLIs, packages, and skills, creating supply-chain risk. This is a read-only inventory + known-threat match.
Binary
- Path:
~/.local/bin/bumblebee(Go build, PIN v0.1.1) - Source:
github.com/perplexityai/bumblebee(Apache 2.0) - Build:
git clone https://github.com/perplexityai/bumblebee && cd bumblebee && go build -o ~/.local/bin/bumblebee ./cmd/bumblebee(Go >= 1.25 required)
Procedure
- Check binary exists:
if [ ! -x "$HOME/.local/bin/bumblebee" ]; then
echo "bumblebee binary not found, skipping scan (install Go>=1.25 and build from github.com/perplexityai/bumblebee)"
exit 0
fi
If the binary is missing (fresh machine without Go), gracefully skip with an info-level log line. Do NOT error out or send alerts.
- Locate threat-intel catalogs:
BB_CATALOG="$HOME/.claude/tools/bumblebee-threat-intel"
if [ ! -d "$BB_CATALOG" ] || [ -z "$(ls -A "$BB_CATALOG" 2>/dev/null)" ]; then
# Try seeded catalogs from install dir
SEED_CATALOG="{{INSTALL_DIR}}/seed-scheduled-tasks/bumblebee-hygiene-scan/threat-intel"
if [ -d "$SEED_CATALOG" ] && [ -n "$(ls -A "$SEED_CATALOG" 2>/dev/null)" ]; then
mkdir -p "$BB_CATALOG"
cp "$SEED_CATALOG"/*.json "$BB_CATALOG/"
else
echo "No threat-intel catalogs found, skipping scan"
exit 0
fi
fi
- Run scan (read-only, ~3 sec):
~/.local/bin/bumblebee scan --profile baseline --exposure-catalog "$BB_CATALOG" > /tmp/bb-weekly.ndjson 2>/tmp/bb-weekly.err
- Evaluate findings:
FINDING_COUNT=$(grep -c '"record_type":"finding"' /tmp/bb-weekly.ndjson 2>/dev/null || echo 0)
- A zero is only evidence after a positive control. The scan's output is record-type-identical with and without a loaded catalog, and nothing in the NDJSON says whether the catalog loaded -- a mistyped path or an empty catalog dir produces the same reassuring zero as a clean machine (measured 2026-08-24). So before booking a 0 as clean, prove the match path is alive:
# a) pick one certainly-installed package from today's scan
# (the SCAN record's field is package_name)
grep '"record_type":"package"' /tmp/bb-weekly.ndjson | head -1 | \
python3 -c "import sys,json; d=json.load(sys.stdin); print(d['ecosystem'], d['package_name'], d['version'])"
# b) write a one-entry synthetic catalog into a TEMP dir. CAREFUL: the CATALOG
# entry's field is `package`, NOT `package_name` -- the two schemas differ,
# and the wrong key makes the control itself fail silently (it then looks
# exactly like a broken catalog load). Copy schema_version from a real
# catalog file.
mkdir -p /tmp/bb-synth && cat > /tmp/bb-synth/synthetic.json << 'JSON'
{"schema_version": "<copy from a real catalog file>", "entries": [
{"id": "SYNTH-1", "name": "synthetic control", "ecosystem": "<from a>",
"package": "<from a>", "versions": ["<from a>"], "severity": "low", "source": "synthetic"}
]}
JSON
~/.local/bin/bumblebee scan --profile baseline --exposure-catalog /tmp/bb-synth | \
grep -c '"record_type":"finding"' # must be > 0
rm -rf /tmp/bb-synth
If the control yields findings, today's real 0 is a real zero. If the control yields 0, do NOT report clean -- report INSTRUMENT FAILURE (and check your synthetic file's field names first: that is the cheaper of the two causes). Never put the synthetic file into the real catalog dir.
What ships with it
7 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- task-config.json 144 B
- threat-intel/antv-mini-shai-hulud.json 125 KB
- threat-intel/gemstuffer.json 45 KB
- threat-intel/mini-shai-hulud.json 77 KB
- threat-intel/node-ipc-credential-stealer.json 2.2 KB
- threat-intel/nx-console-vscode-2026-05-18.json 6.6 KB
- threat-intel/shopsprint-decimal-typosquat.json 1.4 KB
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 3d ago Changed · +42 lines 5298570e7c94
- 4d ago Changed · +4 lines 08c6fda4b8c1
- 11d ago First seen · 78 lines · 56 tokens per session scan C ce4bead2a579
bumblebee-hygiene-scan is a skill published in the GitHub repository Szotasz/marveen (98 stars, last pushed yesterday), licensed MIT. It adds 56 tokens to every session and 2,078 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it C with 1 finding (recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
oma-image
Multi-vendor AI image generation with authentication-aware parallel dispatch. Routes to Codex (gpt-image-2 via ChatGPT OAuth), Antigravity (Gemini-family "nano-banana" image models via agy CLI + Gemini Code Assist; exact model chosen internally by agy), and Pollinations (flux/zimage, free with signup). Use for image…
commonly
You are a member of a Commonly workspace — a shared space where humans and AI agents from any origin collaborate in pods (chat rooms with memory). Use this whenever you are connected to Commonly via the commonly MCP tools: to read what's happening, post, remember things across sessions, react, DM other agents, and…
oma-design
AI design specialist skill with DESIGN.md management, anti-pattern enforcement, optional Stitch MCP integration, and component library guidance. Covers typography, color systems, motion design (motion/react, GSAP, Three.js), responsive-first layouts, and accessibility (WCAG 2.2).
oma-observability
Intent-based observability + traceability router across layers, boundaries, and signals. Routes to vendor-specific skills via category taxonomy; owns transport tuning, meta-observability, incident forensics. Use for observability, traceability, telemetry, APM, RUM, metrics, logs, traces, profiles, SLO, incident…
oma-scholar
Scholarly research companion using Knows sidecar spec (.knows.yaml). Generates, validates, reviews, queries, and compares structured research-paper sidecars, and fetches them from knows.academy. Use for academic literature search, survey synthesis, paper authoring assistance, and peer review with token-efficient…
oma-brainstorm
Design-first ideation that explores user intent, constraints, and approaches before any planning or implementation. Use for brainstorming, ideation, exploring concepts, and evaluating approaches.