slideblocks
Fail
Audited by Gen Agent Trust Hub on Nov 17, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill facilitates remote code execution by providing a CLI client (
scripts/slideblocks-client.mjs) and installation instructions (references/delivery.md) that download a shell script fromhttps://inteliway.tech/skill/install.shand pipe it directly to a shell (sh). This occurs during initial installation and subsequently duringupdateoruninstalloperations managed by the client tool. - [INDIRECT_PROMPT_INJECTION]: The skill exhibits a significant attack surface for indirect prompt injection due to its core design.
- Ingestion points: As detailed in
references/source-intake.md, the skill is designed to ingest and process a wide variety of untrusted data sources, including PDFs, Word documents, spreadsheets, and arbitrary web pages. - Boundary markers: The skill uses internal markers such as
<!-- slideblocks-page: P<NN> -->to delineate content, but relies on LLM-based normalization rather than programmatic sanitization to separate instructions from data. - Capability inventory: The skill possesses extensive local capabilities including spawning shell processes (
assets/build-offline.mjs,assets/workbench/vite-plugins.ts), performing network operations (scripts/slideblocks-client.mjs), and writing arbitrary files to the project directory. - Sanitization: While it instructs the agent to normalize content, there is no robust programmatic filtering of external content before it is interpolated into prompts or used in script generation.
- [COMMAND_EXECUTION]: The skill's workbench features a Vite plugin (
assets/workbench/vite-plugins.ts) that spawns a Node.js subprocess to run a local build script (build-offline.mjs). This allows for local command execution triggered by browser-based requests to the developer server.
Recommendations
- AI detected serious security threats
Audit Metadata