slides-critique

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection vulnerability surface by ingesting external, potentially untrusted slide data and interacting with file-modification tools.
  • Ingestion points: Content is ingested from output.pptx via the slides inspect command and directly from slides.json during Step 1 of the process.
  • Boundary markers: Absent. There are no clear delimiters or instructions telling the model to ignore embedded prompt injections or system commands contained within the presentation text.
  • Capability inventory: The skill possesses capabilities to modify files via the slides edit and slides apply commands which run local system operations based on the model's critique.
  • Sanitization: Absent. No filtering or validation is performed on the parsed text content of the slides before it is interpolated into the agent context.
  • [COMMAND_EXECUTION]: The skill executes multiple local shell commands using the uvx tool manager to run agent-slides for inspecting, editing, and running quality assurance checks on presentation decks (output.pptx). These tools are well aligned with the skill's primary objective of critiquing presentations.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 06:42 AM
Security Audit — agent-trust-hub — slides-critique