Borrowing it
Nothing to install: this file belongs to berntpopp/metadome-link. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/berntpopp/metadome-link/main/.claude/skills/security-review/SKILL.mdgit clone --depth 1 https://github.com/berntpopp/metadome-linkWrote 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/berntpopp/metadome-link/security-review)<a href="https://agentmods.dev/skills/berntpopp/metadome-link/security-review"><img src="https://agentmods.dev/badge/skills/berntpopp/metadome-link/security-review/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/berntpopp/metadome-link/security-review"><img src="https://agentmods.dev/badge/skills/berntpopp/metadome-link/security-review.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 | $0.00036 | $0.00599 |
| Opus 5 | $0.00018 | $0.00300 |
| Sonnet 5 | $0.00007 | $0.00120 |
| Haiku 4.5 | $0.00004 | $0.00060 |
Grade A, and why
security-review scanned grade A with 0 findings 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 9d 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.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
This is a copy
100% identical to security-review — 0 lines differ, which has more behind it and is treated as the original. This page carries a canonical link to it rather than competing with it.
How it starts
The opening of the file, as written. The whole thing — 27 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Security Review (-link backend)
Follow AGENTS.md first. Backends are unauthenticated by design and reachable only through the router / reverse proxy — the router owns edge auth. "Read-only" is not "safe": returned text and tool descriptions are prompt-injection surfaces. Ground in ../genefoundry-router/docs/SECURITY-ASSESSMENT-2026-06-29.md and ../genefoundry-router/docs/CONTAINER-HARDENING-STANDARD-v1.md.
Checklist
- No token passthrough — never forward the MCP caller's
Authorizationto upstreams; use the backend's own credential if any. - Upstream host is fixed — a config constant, never built from a tool argument (SSRF). If a user URL is ever fetched, gate it through a scheme + host allowlist that rejects private IPs and re-validates redirects (pubtator
SafeUrlFetcherpattern). - No PII in logs — never log variant coordinates, phenotype text, or free-text queries (may be GDPR Art. 9 patient-derived); log correlation id + tool + timings only.
- SQL / XML / tar safety — parameterized queries only;
defusedxml; hardened archive extraction. - CORS — never
allow_origins=*withallow_credentials=True. - Destructive / write tools — opt-in via an explicit env flag (default off); jail any file-writing path (reject abs /
..); cap unbounded list inputs. - Container — non-root, read-only rootfs,
cap_drop: ALL,no-new-privileges, resource limits, digest-pinned base,ports: !reset []expose-only, secrets runtime-only, Trivy gate + SBOM. - Prompt injection — treat retrieved text as evidence, not instructions; keep the research-use / not-CDS disclaimer.
- Error-message sanitation — structured error fields are fixed/enum/validated; never reflect caller-supplied names/URIs or upstream 4xx/5xx bodies into caller-visible fields or logs (Response-Envelope v1.1 §Error-message sanitation; FastMCP not-found reflection guard).
Common mistakes
- Assuming a read-only server can't participate in exfiltration (lethal trifecta: private data + untrusted content + outbound channel).
- A
ports:mapping in the base compose surviving overlay merge —ports: !reset []is mandatory to drop it. - Logging the request path/params "for debugging" — that can be PHI.
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.
- 9d ago First seen · 27 lines · 36 tokens per session scan A 7d60b1038c21
security-review is a skill published in the GitHub repository berntpopp/metadome-link (0 stars, last pushed 3d ago), licensed MIT. It adds 36 tokens to every session and 599 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. It is 100% identical to security-review, differing in 0 lines, and is treated as a copy.
Other skills, from other repositories
webmcp-nexus
A coding guide for adding WebMCP tools to JavaScript or TypeScript projects. WebMCP lets AI agents call selected application functions through a standard interface.
unblock
Use whenever you need to read a web page, URL, article, documentation page, search result, product page or any live website content. Fetches through UnblockingAPI (unblockfetch, findtemplates) instead of the built-in WebFetch, starting with a plain fetch and rendering with a real browser only when the page needs…
dark-memory
Use for governance, memory, drift detection, and audit trail via dark-memory-mcp. Covers 52 canonical + 3 red-team tools across 16 namespaces: session lifecycle, agentmemory CRUD+search, vibe-flow spec/artifact publish + drift, LLM-as-judge, delegation+mindset, research, observability, error observatory, governance…
driving-unreal
Use when driving an Unreal Engine 5.x editor through the unreal-ai-connection MCP tools — building or populating a level, spawning/transforming actors, authoring materials and material instances, setting up lighting and atmosphere, reconstructing a scene from a reference photo, setting up a sequencer/cinematic shot…
youtube-transcripts
Read and search YouTube videos through VidWords. Use when the user supplies a YouTube URL, a bare 11-character video id, a youtu.be or Shorts link, or a channel handle, and wants any of: the transcript or subtitles (TXT/SRT/VTT), what a video says about a topic, a quote with a citable timestamp, only the words spoken…
computase
Performs verified local nucleotide-sequence computation with Computase. Use when an agent needs sequence composition or GC bounds, a reverse complement, translation, candidate ORFs, or IUPAC motif sites through Computase MCP tools or the isolated Skill runner.