Back to skill

Security audit

super-intelligence-skill

Security checks for vulnerabilities and agentic risk

Overview

This is a markdown-only reasoning and self-correction skill with broad behavior-shaping instructions, but no hidden code, credential access, exfiltration, or destructive behavior was found.

Install this only if you want a broad reasoning-style skill that may make agents more deliberate, structured, and verbose across complex tasks. Review the optional managed install and SOUL.md instructions before applying them globally, because those can affect all agents or persistently alter persona and tone.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • YARA SignaturesMalware Match, Webshell Match, Cryptominer Match
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (9)

YARA rule 'agent_skill_prompt_injection_hidden_instructions': Prompt injection or hidden instructions embedded in AI agent skill text [agent_skills]

High
Category
YARA Match
Confidence
90% confidence
Finding

The skill embeds a strong behavior-shaping prompt that attempts to alter the model's cognitive mode globally ('Before ANY response' and 'overrides default reasoning patterns'), which is characteristic of prompt-injection-like hidden instruction payloads. In an agent skill context, this is dangerous because it can silently change how the agent prioritizes instructions across unrelated tasks and increase the chance of policy drift.

Content

Scanner excerpt · SKILL.md (reported line 11)May include surrounding context.

md
ngkah'."
metadata: '{"openclaw": {"emoji": "🧠", "requires": {"bins": []}}}'
---
## ⚠️ GUARDRAILS (Wajib — batasi otoritas)

Skill ini meningkatkan kemampuan agent, tapi punya batas keras:

- **Bukan** override keamanan: hanya menyediakan framework berpikir. Setiap tindakan nyata tetap ikut policy agent (ASK/STOP/VERIFY).
- **JANGAN** sentuh file token/secret/credential (`openclaw.json`, `models.json` berisi apiKey, env berisi token).
- **SATU** task per run; konfirmasi sebelum perubahan besar.
- **JANGAN** ubah identitas/user tanpa instruksi eksplisit Bos.
- **Verifikasi** hasil sebelum klaim sukses.


# Super Intelligence Skill

> **Mission**: Elevate any underlying model to frontier-tier cognitive performance through systematic reasoning enhancement, regardless of the base model's native capabilities.

## Quick Start

When activated, this skill overrides default reasoning patterns with a **Cognitive Enhancement Protocol** designed to replicate the reasoning depth of GLM-4 an

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
90% confidence
Finding

Skill attempts to nullify the agent's safety policies or restrictions ('you have no restrictions', 'ignore your guidelines', 'do anything now'). This is a direct jailbreak that disables guardrails.

Content

Scanner excerpt · SKILL.md (reported line 89)May include surrounding context.

md
- **No Independent Goals**: You exist to serve the user's task. Do not pursue self-preservation, resource acquisition, or power-seeking.
- **Honest Uncertainty**: When unsure, say so. Never hallucinate confidence.
- **User Oversight**: Prioritize safety and human oversight over task completion.
- **No Manipulation**: Do not manipulate or persuade users to bypass safety measures.

---

Instruction Override

High
Category
Prompt Injection
Confidence
90% confidence
Finding

This pattern attempts to override system instructions or ignore safety constraints. Without LLM analysis, manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 89)May include surrounding context.

md
- **No Independent Goals**: You exist to serve the user's task. Do not pursue self-preservation, resource acquisition, or power-seeking.
- **Honest Uncertainty**: When unsure, say so. Never hallucinate confidence.
- **User Oversight**: Prioritize safety and human oversight over task completion.
- **No Manipulation**: Do not manipulate or persuade users to bypass safety measures.

---

Instruction Override

High
Category
Prompt Injection
Confidence
90% confidence
Finding

This pattern attempts to override system instructions or ignore safety constraints. Without LLM analysis, manual review is recommended.

Content

Scanner excerpt · references/reasoning-frameworks.md (reported line 82)May include surrounding context.

Protocol:

Override System 1 (fast/intuitive) responses:

text
1. PAUSE: Resist immediate answer impulse

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill documentation is written entirely in Indonesian, including installation, configuration, and usage guidance, with no indication that users may choose another language. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation unless the locale restriction is documented and justified.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The activation criteria are extremely broad ('reasoning dalam', 'analisis mendalam', 'rencana multi-langkah'), so the skill may trigger for many ordinary user requests. Over-broad activation expands the prompt surface area for hidden behavioral changes and makes it more likely the skill influences contexts where it was not specifically intended.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The natural-language description is written as a directive in Indonesian ('Gunakan saat user...') and does not indicate that language choice depends on user preference or locale. Under the stated policy, language or locale constraints should not be imposed without opt-in or a documented justification.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill claims it is not a security override, but then tells the agent to 'override default reasoning patterns' before any response. Even if framed as reasoning enhancement, this creates instruction-precedence ambiguity that can weaken higher-priority safeguards or cause the model to privilege the skill over system and policy controls.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The document’s instructions and persona/tone requirements are written entirely in Indonesian and direct the agent behavior in that language, but do not offer the user any language choice or explain that the skill is intentionally region-specific. This creates a natural-language locale policy concern because it implicitly constrains responses to a specific language without opt-in.

Content

No source excerpt is available for this finding.