dart-dsd
Pass
Audited by Gen Agent Trust Hub on Nov 18, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill provides instructions for the agent to download and install the
dsdCLI tool from the author's public GitHub repository (https://github.com/cpaikr/dsdai/releases). These instructions are transparent and emphasize using official releases and checksum verification. - [COMMAND_EXECUTION]: The skill relies on two command-line tools: the
dsdCLI for document processing and thegh(GitHub) CLI for reporting issues. Both are standard for developer-oriented tools, and the instructions require explicit user approval before performing actions like filing GitHub issues. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external
.dsdfiles which may contain untrusted data. - Ingestion points: Data enters the agent's context when reading
.dsdfiles (specificallycontents.xml) through thedsdCLI tool as described inSKILL.mdandreferences/reading.md. - Boundary markers: The skill uses a dedicated CLI tool that returns structured JSON or formatted text, providing a layer of abstraction between the raw file content and the agent's prompt.
- Capability inventory: The skill possesses capabilities for file system read/write and execution of the
dsdandghcommand-line tools. - Sanitization: The instructions mandate quoting figures and text directly from the tool's output and require post-edit verification using
dsd checkanddsd diffto ensure no unintended changes were made. - [DATA_EXFILTRATION]: The skill explicitly instructs the agent to treat documents as confidential, process them locally, and avoid uploading them to external services or shared locations.
Audit Metadata