slides-polish
Pass
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill mandates the creation and execution of a local Python script (
inspect_pptx.py) at runtime. It defines a specific 'contract' (schema and logic) for this script and instructs the agent to implement and run it if the file is not found in the working directory. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted slide content (text, titles, and layout data) which is then interpolated into prompts for a high-reasoning sub-agent (
spawn_agent). - Ingestion points: Slide data is extracted from user-provided
.pptxand.texfiles during Phase 0 and Phase 1. - Boundary markers: The skill mentions that text content in the inspector output is 'escaped', providing a basic level of boundary separation in the generated JSON metadata.
- Capability inventory: The skill has access to
Write,Edit, andBashtools, allowing it to modify files and execute commands based on the reviewer agent's output. - Sanitization: Beyond JSON escaping of extracted metadata, there is no evidence of robust sanitization or filtering of the slide content before it is processed by the sub-agent.
- [COMMAND_EXECUTION]: The workflow relies on several external system utilities for slide rendering and PDF processing, including
pdftoppm,soffice,xelatex,latexmk, andmutool. These commands are executed via theBashtool to facilitate the visual review process.
Audit Metadata