shared-content-references
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_CONTEXT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of instructional prose from external files into the agent context using path variables like
${CLAUDE_PLUGIN_ROOT}. Ingestion points include references inSKILL.mdandreferences/annotation-format.md. While the skill suggests descriptive link annotations, it lacks technical boundary markers or sanitization for the loaded content. Capabilities include shell execution and file reading via the platform's path resolution features. This pattern is fundamental to the skill's purpose of sharing instructions. - [DYNAMIC_CONTEXT_INJECTION]: The skill documentation describes the platform's dynamic context injection syntax, providing examples such as
!find ${CLAUDE_PLUGIN_ROOT}/docs -name '*.md' -type f | sortto list files at load time. These examples demonstrate legitimate environment inspection for development purposes. - [COMMAND_EXECUTION]: The verification guidelines in
references/verification.mdinstruct users to runuvx skilllint@latest check <plugin-path>, a command that executes an external linting utility to ensure skill integrity.
Audit Metadata