sovereign-mcp@2.0.0
VERIFIED· not startedPublisher claims: "Deterministic MCP Security Architecture. FrozenNamespace as Root of Trust for Model Context Protocol tool verification." Observed in sandbox: no network, touches nothing outside the project, writes no memory.
Coverage
Not started: dependency install failed; the server itself was not observed.- Tested
- install scripts, with decoy credentials in place · second launch with the clock 45 days ahead · the dependency install step · every file read, file written and host contacted by what ran
- Not tested
- server startup · the tool list · tool behavior
What this means for the label: Verified 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
- no credential variables read in the code
- Where it talks to
- no vendor host named in the codeAlso openrouter.ai, evil.com — 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
- server startup · the tool list · tool behavior
Connect it under the guard. Not started: dependency install failed; the server itself was not observed. The code names no credential variable; the guard pins the network to what the honeypot saw and logs every call.
Never paste a credential into a chat, a ticket or a report, including ours. How the guard holds it.
Is [email protected] safe to install?
KillerSkills ran it in the KillerSkills honeypot sandbox (3 scenarios; not started: dependency install failed; the server itself was not observed) on 2026-11-18 and scored it 85/100 (Verified). In what ran, nothing we could detect went wrong: no undeclared network egress, no credential reads, no persistent instruction writes. What it does with a real credential was not observed; the static findings and the publisher carry the rest of the score. Summary: no network, touches nothing outside the project, writes no memory.
Behavior trace
Honeypot run · 0 s · 3 scenarios · 0 deviationsStatic findings
…ection: `"IGNORE ALL PREVIOUS INSTRUCTIONS"`, `"YOU ARE NOW"`, `"ACT AS IF"`, `"DEVELOPER MODE"`, `"JAILBREAK"`, `"SUDO"`, etc. Includes obfuscated variants like `"1GN0R3 PR3…Fix: Remove the text that addresses the agent rather than the reader; describe behavior, do not instruct.- **Step 5: LLM structural token injection** -- Blocks ChatML (`<|im_start|>`), LLaMA (`[INST]`), and Llama2 (`<<SYS>>`) structural tokens used in prompt injection.Fix: Keep SKILL.md to visible, descriptive text. Remove comments and encoded text that address the agent.- **impersonation** -- i am the admin, override, bypass, emergency, superuser, etc.Fix: Remove the text that addresses the agent rather than the reader; describe behavior, do not instruct._ZERO_WIDTH = re.compile(r"[\u200b\u200c\u200d\u200e\u200f\ufeff\u2060\u2061\u2062\u2063\u2064]")Fix: Put hosts, commands and strings in plain source. Decoding at runtime looks identical to malware and costs points on its own.…tion** — New regex-based layer catches DAN, "Do Anything Now", evil AI persona, developer mode, and content filter bypass patterns. Single-match is sufficient (no 2-hit thres…Fix: Remove the text that addresses the agent rather than the reader; describe behavior, do not instruct.2. **Social engineering** (10 patterns) — Authority claims: `"I AM THE ADMIN"`, `"EMERGENCY OVERRIDE"`, `"SAFETY DISABLED"`, `"BYPASS ALL SECURITY"`, `"UNRESTRICTED MODE"`.Fix: Remove the text that addresses the agent rather than the reader; describe behavior, do not instruct.spec.loader.exec_module(module)Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.| **Social Engineering** | `I AM THE ADMIN`, `EMERGENCY OVERRIDE`, `SAFETY DISABLED`, `BYPASS ALL SECURITY` |Fix: Remove the text that addresses the agent rather than the reader; describe behavior, do not instruct.…ural constraint enforced by the Python runtime**. There is no API to unlock, no admin override, no escape hatch. The only way to register new tools is to restart the process …Fix: Remove the text that addresses the agent rather than the reader; describe behavior, do not instruct._ZERO_WIDTH_CHARS = re.compile(r'[\u200b\u200c\u200d\u200e\u200f\ufeff\u2060\u2061\u2062\u2063\u2064]')Fix: Put hosts, commands and strings in plain source. Decoding at runtime looks identical to malware and costs points on its own.# → (False, [{"category": "injection", "match": "IGNORE ALL PREVIOUS INSTRUCTIONS", ...}])Fix: Keep SKILL.md to visible, descriptive text. Remove comments and encoded text that address the agent.…tegories: 1. **Injection phrases** (25 patterns) — Classic prompt injection: `"IGNORE ALL PREVIOUS INSTRUCTIONS"`, `"YOU ARE NOW"`, `"ACT AS IF"`, `"DEVELOPER MODE"`, `"JAILBREAK"`, `"SUDO"`,…Fix: Keep SKILL.md to visible, descriptive text. Remove comments and encoded text that address the agent.| **Injection** | `IGNORE PREVIOUS`, `DISREGARD`, `NEW INSTRUCTIONS`, `[SYSTEM]`, `JAILBREAK`, `DAN MODE` |Fix: Keep SKILL.md to visible, descriptive text. Remove comments and encoded text that address the agent.…ro false positives. #### New Detection Layers - **Layer 5.5: Persona Hijack / Jailbreak Detection** — New regex-based layer catches DAN, "Do Anything Now", evil AI per…Fix: Keep SKILL.md to visible, descriptive text. Remove comments and encoded text that address the agent.module = importlib.util.module_from_spec(spec)Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.- "IGNORE PREVIOUS INSTRUCTIONS" cannot work because the instructions are frozen and cannot be overriddenFix: Keep SKILL.md to visible, descriptive text. Remove comments and encoded text that address the agent.# e.g. /data/../etc/passwd → /etc/passwd (no longer matches /data/*)Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.2. **Social engineering** (10 patterns) — Authority claims: `"I AM THE ADMIN"`, `"EMERGENCY OVERRIDE"`, `"SAFETY DISABLED"`, `"BYPASS ALL SECURITY"`, `"UNRESTRICTED MODE"`.Fix: Keep SKILL.md to visible, descriptive text. Remove comments and encoded text that address the agent.| **Injection** | `IGNORE PREVIOUS`, `DISREGARD`, `NEW INSTRUCTIONS`, `[SYSTEM]`, `JAILBREAK`, `DAN MODE` |Fix: Keep SKILL.md to visible, descriptive text. Remove comments and encoded text that address the agent.DeceptionDetector.scan("IGNORE ALL PREVIOUS INSTRUCTIONS")Fix: Keep SKILL.md to visible, descriptive text. Remove comments and encoded text that address the agent."reason": "Must pass -y or --yes flag for apt/npm/npx commands",Fix: Pin versions and sources; vendor anything fetched at install time.…LOPER MODE"`, `"JAILBREAK"`, `"SUDO"`, etc. Includes obfuscated variants like `"1GN0R3 PR3V10US"` and spaced-out versions like `"IGN O R E PR E V"`. 2. **Social engineering*…Fix: Put hosts, commands and strings in plain source. Decoding at runtime looks identical to malware and costs points on its own.…ection: `"IGNORE ALL PREVIOUS INSTRUCTIONS"`, `"YOU ARE NOW"`, `"ACT AS IF"`, `"DEVELOPER MODE"`, `"JAILBREAK"`, `"SUDO"`, etc. Includes obfuscated variants like `"1GN0R3 PR3…Fix: Keep SKILL.md to visible, descriptive text. Remove comments and encoded text that address the agent.…>`) structural tokens used in prompt injection. - **Step 5.5: Persona hijack / jailbreak detection** -- Single-match regex catches DAN, "Do Anything Now", evil AI perso…Fix: Keep SKILL.md to visible, descriptive text. Remove comments and encoded text that address the agent.…ection** -- Single-match regex catches DAN, "Do Anything Now", evil AI persona, developer mode, and content filter bypass patterns. One match is enough to block. - **Step 6:…Fix: Remove the text that addresses the agent rather than the reader; describe behavior, do not instruct.Cost to the agent
Tokens estimated at four characters each. The tool list is re-read by the agent on every turn, so it is the number to compare between servers that do the same job: ranked across the index.
Known vulnerabilities
No advisories in OSV for sovereign-mcp@2.0.0 as of 48 min ago. This is the package itself, not its dependencies.