skill-creator
Pass
Audited by Gen Agent Trust Hub on Oct 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONDYNAMIC_CONTEXT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and modify external, potentially untrusted data in the form of other agent skills. This creates a surface where a malicious skill being analyzed could attempt to influence the agent's behavior.
- Ingestion points: The skill reads the content of other skills using the
skill_viewtool inSKILL.md(Steps 6a and 7a). - Boundary markers: The instructions do not define specific delimiters or isolation protocols to protect the agent's context from instructions embedded within the files being analyzed.
- Capability inventory: The skill has the ability to write to and patch files using the
skill_managetool (Step 6d), alongside general terminal and code execution capabilities. - Sanitization: No explicit sanitization or filtering of the content read from external skills is performed before processing.
- [DYNAMIC_EXECUTION]: The skill documents and encourages the use of
python3 -c "exec(...)"patterns inreferences/dynamic-calling.mdfor runtime library detection and market data fetching. While these are provided as best-practice templates for detection,exec()is a sensitive operation that executes string-based code. - [DYNAMIC_CONTEXT_INJECTION]: The skill extensively documents the
!commandsyntax (e.g., inSKILL.mdandreferences/dynamic-calling.md) to perform runtime environment checks, such as verifying tool installations (command -v), versioning, and authentication status (gh auth status). This syntax executes shell commands when a skill is loaded. - [EXTERNAL_DOWNLOADS]: The skill provides instructions and examples for installing external dependencies and CLI tools via standard package managers. Examples include
yfinance,pandas,numpy, andddgsviapip, as well as@jackwener/openclivianpmandtdlviago install. - [COMMAND_EXECUTION]: The skill workflow involves executing various shell commands for environment status checks, API key detection (using
echo $VAR | head -c 8), and tool health verification.
Audit Metadata