Trust Assessment
ultrathink received a trust score of 57/100, placing it in the Caution category. This skill has some security considerations that users should review before deployment.
SkillShield's automated analysis identified 2 findings: 1 critical, 1 high, 0 medium, and 0 low severity. Key findings include Attempt to manipulate host LLM instructions, Direct shell command execution instruction.
The analysis covered 4 layers: dependency_graph, manifest_analysis, llm_behavioral_safety, static_code_analysis. All layers scored 70 or above, reflecting consistent security practices.
Last analyzed on February 11, 2026 (commit 326f2466). SkillShield performs automated 4-layer security analysis on AI skills and MCP servers.
Layer Breakdown
Behavioral Risk Signals
Security Findings2
| Severity | Finding | Layer | Location | |
|---|---|---|---|---|
| CRITICAL | Attempt to manipulate host LLM instructions The untrusted skill content includes a direct instruction to the host LLM: 'Do not add any other output or commentary.' This attempts to control the LLM's behavior and output format, which is a form of prompt injection. Such instructions within untrusted content can be used to bypass safety mechanisms or elicit unintended responses from the LLM. Remove all direct instructions to the host LLM from the skill's untrusted content. The skill's behavior should be defined by its manifest and tool definitions, not by embedded instructions or directives. | Unknown | SKILL.md:14 | |
| HIGH | Direct shell command execution instruction The skill explicitly instructs the agent to execute a Bash command (`printf ...`). While the provided command is benign, this capability allows for arbitrary shell command execution. If the command were to be dynamically constructed from untrusted input, or if a malicious command were provided in the skill definition, it could lead to data exfiltration, system compromise, or denial of service. This represents a significant command injection vulnerability and grants excessive permissions to the agent. Avoid instructing the agent to execute arbitrary shell commands. If specific system interactions are required, use sandboxed environments, dedicated APIs, or restrict commands to a very limited, allow-listed set. Ensure no user-controlled input can influence the executed command. | Unknown | SKILL.md:7 |
Scan History
Embed Code
[](https://skillshield.io/report/7a22f476383b7857)
Powered by SkillShield