presentation-design
Pass
Audited by Gen Agent Trust Hub on Apr 4, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands (
ls) to inspect local directories for installed skills on first activation. - [DATA_EXFILTRATION]: The skill performs directory listing of environment-specific paths (e.g.,
~/.claude/skills/,~/.agent/skills/) to fingerprint the local installation state and identify missing companion skills. - [EXTERNAL_DOWNLOADS]: The skill encourages the installation of additional packages via the
npx skills addcommand, which fetches code from external vendor repositories. - [PROMPT_INJECTION]: The skill ingests untrusted user content for presentation design and structuring without using boundary markers or sanitization, presenting a vulnerability to indirect prompt injection.
- Ingestion points: User-supplied outlines, data, and presentation goals in
SKILL.md. - Boundary markers: Absent.
- Capability inventory: File system listing (
ls) and skill installation vianpx. - Sanitization: None identified in the instruction set.
Audit Metadata