ac-generating-slides
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses a Python command-line wrapper (
scripts/cli.py) to execute themarputility and system file openers. The script constructs these commands using argument lists rather than raw shell strings, which mitigates the risk of command injection. - [INDIRECT_PROMPT_INJECTION]: The skill's workflow involves reading external 'branding skills' and processing user-driven dialogue to generate presentation content. This ingestion of external data constitutes a surface for indirect instructions, although the risk is limited by the skill's specific output format (Markdown slides).
- Ingestion points: Loading branding configuration files from agent memory and processing user input during the authoring dialogue.
- Boundary markers: No explicit markers are defined to isolate instructions within branding files.
- Capability inventory: File writing (Markdown decks) and subprocess execution (Marp PDF rendering).
- Sanitization: The skill does not specify explicit sanitization or filtering for external branding content.
- [SAFE]: No obfuscation, persistence mechanisms, or unauthorized network operations were identified. The external dependencies and tools referenced (Marp, Chromium) are well-known and standard for the skill's stated purpose.
Audit Metadata