T01 · Skill Instruction Hijacking
- Location
tool.sh:29- Finding
Untrusted Search Results Are Embedded Directly into an LLM Prompt
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This skill performs the described local SearXNG drug-name lookup, but users should treat medication queries and returned search snippets as sensitive and untrusted.
Install only if you are comfortable sending queried drug names to the local SearXNG service and any upstream engines it uses. Run the LLM analysis with normal caution: ignore instructions found inside search results, and prefer authoritative medical sources for decisions.
tool.sh:29Untrusted Search Results Are Embedded Directly into an LLM Prompt
tool.sh:23Unsafe String Interpolation Produces Malformed or Manipulable JSON Output
The skill explicitly sends user-provided drug names to a local HTTP SearXNG service and returns raw search context for LLM analysis, but the documentation does not warn users that their input will be transmitted or that untrusted search content will be fed back into the model. This creates a real privacy and prompt-injection risk, especially for sensitive medication queries that may reveal health information and for raw search snippets that can contain adversarial instructions.
The natural-language strings in the script, including errors, notes, and the suggested LLM analysis prompt, are all fixed in Chinese with no option for another language or locale. This creates a language/locale policy concern because the skill imposes a specific language without user opt-in or documented justification.
The README states that the skill sends the queried drug name to a local HTTP SearXNG service but does not clearly warn users that their input will be transmitted over a network interface. Even though the target is localhost, this can still expose potentially sensitive medication queries to a locally running service, its logs, or any process/user controlling that endpoint, creating a privacy and transparency issue.
The file's natural-language instructions and examples are entirely in Chinese, which can amount to a language-policy issue when the skill does not state that it is intentionally limited to Chinese-speaking users or offer an alternative language. The policy allows locale constraints when they are explicit and justified, but that justification is not present here.
This shell script transmits the user-supplied drug name to a SearXNG HTTP endpoint via curl. Although comments describe the behavior for developers, there is no user-facing prompt, warning, or visible disclosure before the network call, so users are not explicitly informed that their input will be sent to a search service.