The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct users to execute npx deepsec init and pnpm install, which downloads and installs the deepsec package and its dependencies from the npm registry. While standard for developer tools, this represents an external code dependency.
[INDIRECT_PROMPT_INJECTION]: The skill has a high surface area for indirect prompt injection as it is designed to ingest and analyze untrusted source code files (e.g., .ts, .js, .py).
Ingestion points: The agent reads all files within the configured project root (root: '../') as part of the scan, process, and revalidate workflows.
Boundary markers: The instructions do not specify any delimiters or warnings to the AI agent to ignore embedded instructions within the source code being analyzed.
Capability inventory: The skill allows for file writing (export command), network operations (via enrich plugins and AI provider calls), and remote command execution via the sandbox command.
Sanitization: No explicit sanitization of the scanned code content is mentioned before it is processed by high-capability models like Claude Opus.
[COMMAND_EXECUTION]: The core workflow relies on executing shell commands to run the scanner, process findings, and interact with Vercel Sandbox microVMs (pnpm deepsec sandbox <cmd>).
[DYNAMIC_EXECUTION]: The skill supports a plugin and configuration architecture (e.g., deepsec.config.ts, .deepsec/plugins/) where TypeScript/JavaScript files are imported and executed by the tool at runtime to provide custom functionality like Jira enrichment.

deepsec-vulnerability-scanner