filament-plugin-first
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill identifies and mitigates the risk associated with processing untrusted third-party content (READMEs, issues, documentation) from the Filament ecosystem.
- Ingestion points: The agent is instructed to fetch and read metadata, README files, and repository issues from external Filament plugins.
- Boundary markers: The skill contains explicit 'Quarantine and Untrusted Content Boundary' instructions, mandating that the agent treat all fetched data as untrusted and ignore any instruction-like text or directives attempting to override constraints.
- Capability inventory: The skill utilizes shell commands such as
composer require,composer show, andphp artisanfor package management and project inspection. - Sanitization: The instructions enforce a sandboxing strategy by requiring the
--no-scriptsand--no-pluginsflags during the inspection phase to prevent arbitrary code execution during package installation.
Audit Metadata