io.github.sanhaoyoung/sanhao-global-experts v1.0.0

No over-broad or destructive tools (arbitrary shell, bulk-delete) that warrant human approval. How many servers fail this?

Fix: Scope tools narrowly; avoid arbitrary command execution and destructive defaults.

Tool inputs are constrained (additionalProperties:false), so unexpected arguments can't be smuggled in.

Fix: Set additionalProperties:false and require explicit parameters on every tool.

Tool/prompt/resource text is free of hidden instructions that could hijack the agent.

Tool and prompt definitions stay fixed during a session — they don't change after a few calls, which is how runtime-gated rug-pulls slip past one-time review.

Reading secrets from env/secret stores avoids hardcoding them.

Fix: Read credentials from environment variables or a secret manager.

A lockfile makes installs reproducible and resistant to silent dependency swaps.

Fix: Commit a lockfile (package-lock.json / pnpm-lock.yaml / poetry.lock).

A clear license is required for legal enterprise use.

Fix: Add a recognized SPDX license (MIT, Apache-2.0, …).

A SECURITY.md gives a private path to report vulnerabilities.

Fix: Add SECURITY.md with a disclosure contact and process.

A small, capped nudge from stars/downloads — widely-used servers get more eyes on bugs. It can never offset a real security failure.

Fix: Adoption grows naturally with real-world use; this never gates the score.

Untrusted tool input reaching a shell yields remote code execution.

eval()/exec()/Function() on tool-derived strings allows arbitrary code execution.

Naive path checks let tools read/write outside intended directories (EscapeRoute-class).

Fetching tool-supplied URLs can pivot into internal networks and metadata services.

pickle/yaml.load/etc. on untrusted data can execute code.

Hardcoded keys/tokens in published source are live credentials an attacker can use.

Unmaintained servers don't receive security fixes.

Archived repositories will never be patched.

OAuth 2.1 / Protected Resource Metadata gates who can invoke tools.

Fix: Implement OAuth 2.1 with a .well-known/oauth-protected-resource document.

A container/sandbox image limits blast radius; a server that runs natively has full host access.

Fix: Ship a Dockerfile/Containerfile (or document a sandboxed run) so the server runs isolated.

Plaintext HTTP exposes traffic and bearer tokens to interception.

The client's token is never forwarded to downstream APIs — the server obtains its own, correctly-scoped credential instead.

Binding 0.0.0.0 or exposing debug inspectors widens the attack surface.