dexter

Fail

Audited by Gen Agent Trust Hub on Oct 26, 2026

Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill clones the entire source code for the research agent from a third-party GitHub repository (https://github.com/virattt/dexter.git) during the installation phase.
  • [REMOTE_CODE_EXECUTION]: After cloning the external repository, the skill executes remote code by running 'bun install' to fetch third-party dependencies and 'bun start' or 'bun query.ts' to execute the downloaded application logic.
  • [DYNAMIC_EXECUTION]: The skill uses 'sed' to programmatically patch the application's source code (tool-executor.ts) at runtime and dynamically generates a new TypeScript runner file (query.ts) using shell redirection.
  • [COMMAND_EXECUTION]: The skill performs multiple shell operations including file system modifications (mkdir, cat, sed), environment configuration, and process management through 'tmux' to run the research agent.
  • [INDIRECT_PROMPT_INJECTION]:
  • Ingestion points: The skill is designed to ingest and process data from external sources including the Financial Datasets API, SEC filings, and web search results from Tavily.
  • Boundary markers: No explicit instruction boundary markers or "ignore embedded instructions" warnings are documented for the data processing pipeline.
  • Capability inventory: The skill has the capability to execute shell commands ('bun'), write files to the workspace, and perform network requests to various APIs.
  • Sanitization: There is no evidence of sanitization or validation of the content retrieved from external financial data sources or web searches before it is processed by the agent.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 26, 2026, 09:37 AM