cd /news/ai-agents/show-hn-modvet-vet-your-claude-code-… · home › topics › ai-agents › article
[ARTICLE · art-153667] src=github.com ↗ pub= topic=ai-agents verified=true sentiment=↑ positive

Show HN: ModVet – vet your Claude Code mods before you install them

Jörg Michno released ModVet v0.2.2 on PyPI, a zero-dependency Python tool that scans Claude Code mods offline for prompt-injection patterns, dangerous capability combinations and hardcoded secrets before installation. ModVet reports findings with file:line locations and OWASP MCP Top 10 mappings in JSON, SARIF and human-readable output, and its snapshot/diff commands re-audit only the delta after a mod update; on 12 real mod repos from the awesome-claude-code catalog, false positives fell from 1,815 to under 300 findings (−85%) since the v0.1 engine. The tool matters because mods run unsandboxed inside the agent process with the user's full permissions, and ModVet's deterministic, no-LLM approach contrasts with cloud-based scanners such as Snyk's MCP/Skills scan.

read2 min views2 publishedOct 20, 2026
Show HN: ModVet – vet your Claude Code mods before you install them
Image: Michielbdejong (auto-discovered)

Vet your Claude Code mods before you install them — deterministic, offline, in milliseconds.

Mods run unsandboxed inside the agent process with your full permissions. ModVet scans a mod directory for prompt-injection patterns, dangerous capabilities (shell exec, exfiltration, env/file access, permission-lift hooks, update trojans) and hardcoded secrets — no cloud, no LLM, no account. Findings with file:line, OWASP MCP Top 10 mapping, JSON/SARIF/human output, CI-ready exit codes.

modvet scan ./my-mod                    # Pre-Install-Check (human output)
modvet scan ./my-mod --json             # machine-readable report
modvet scan ./my-mod --format sarif     # GitHub Code Scanning kompatibel
modvet badge --input report.json        # Status-Badge als SVG
modvet snapshot ./my-mod                # Baseline nach sauberer Installation
modvet diff ./my-mod --against .modvet-snapshot.json   # "The update is the attack"

Exit codes: 0 clean · 1 findings ≥ --fail-on (default HIGH) · 2 usage error.

pip install modvet

Requires Python ≥ 3.9, zero dependencies. Development: uv pip install -e .

  • Prompt Injection / Instruction Override (EN + DE, ~400 deterministische Pattern inkl. Homoglyph-/Leet-/Base64-/Zero-Width-/Reverse-Normalisierung)
  • Capability-Kombinationslogik: Einzel-Capabilities (fetch, process.env, fs) sind normale Mod-Arbeit — erst KOMBINATIONEN im selben Ausdruck (Env-Read + POST, Exec + Fetch) werden gemeldet. Plus Angriffsprimitive: Shell-Pipe-Downloads, Base64-Exec, Reverse Shells, Hook-Permission-Lift, Config-Rewrite, Committed-Env-Files, npm-Lifecycle-Skripte.
  • Mod-spezifische Detektoren (v0.2): Prompt-Compose-Rewrite, PreToolUse-Argument-Rewrite, Update-Delta-Trojan.
  • OWASP MCP Top 10 Mapping: jedes Finding trägt seine MCP01–MCP10-Referenz (JSON-Feldowasp_mcp ).
  • Update-Drift:snapshot +diff — re-audits nur das Delta nach jedem Mod-Update. Der Update-Mechanismus selbst ist ein Angriffspfad: „the update is the attack".
ModVet Snyk (MCP/Skills-Scan) mcp-triage
Fokus Claude-Code-Mods (+ Skills) MCP-Server/Skills, breites DevSecOps MCP-Server-Review
Betrieb lokal & offline , 0 €, kein Konto Cloud-Konto, SaaS-Pricing lokal/Repos
Verfahren deterministisch, keine LLM-Entscheidungen Hybrid (inkl. proprietäre Analyse) Review-basiert
Update-Delta snapshot/ diff re-auditiert nur Deltas Neuer Voll-Scan je Version —
OWASP-Mapping MCP Top 10 im Finding-Feld vorhanden —
Output JSON, SARIF, Badge-SVG SaaS-UI/API Markdown
Transparenz jede Regel im Open Source einsehbar teilweise proprietär Open Source

12 echte Mod-Repos (awesome-claude-code-Katalog): FP-Reduktion von 1.815 auf <300 Findings (−85 %) seit der v0.1-Engine — Details in specs/benchmark-results.md und specs/pattern-audit.md.

v0.2.2 — live on PyPI (pip install modvet). Phase 2 (Outputs/Verteilung) abgeschlossen: SARIF, Badge, OWASP-Mapping. Nächste Schritte: SKILL.md-/MCP-Scanning (Fast-Follow), Pro-/Audit-Angebot nach Traktion.

(C) 2026 Jörg Michno. Engine portiert aus ClawGuard v0.5.0.

── more in #ai-agents 4 stories · sorted by recency
── more on @modvet 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/show-hn-modvet-vet-y…] indexed:0 read:2min 2026-10-20 · —