load-template
Fail
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- Remote Code Execution (HIGH): The skill executes a remote script via piped curl to bash from an untrusted GitHub repository (
tkarakai/ai-agent-instruction-templates). This allows for arbitrary code execution on the user's system. - External Downloads (MEDIUM): The skill downloads instruction templates from a repository not listed as a trusted external source, bypassing standard security reviews.
- Prompt Injection (LOW): The skill presents an attack surface for indirect prompt injection.
- Ingestion points: Template files such as
AGENTS.mdandtemplate.yamlare downloaded into the.agents/directory. - Boundary markers: Absent. No delimiters or warnings are provided to the agent to disregard instructions within the downloaded content.
- Capability inventory: The execution block in
SKILL.mdfacilitates shell execution (bash) and network access (curl). - Sanitization: Absent. There is no evidence of validation or sanitization of the downloaded template content before processing.
Recommendations
- AI detected serious security threats
Audit Metadata