The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides functions to execute shell commands within managed sandbox environments through methods such as sandbox.process.exec. Evidence is found in SKILL.md and SDK reference files.
[DYNAMIC_EXECUTION]: Core functionality involves the creation and management of dynamic compute environments (MicroVMs) for running code, development servers, and batch jobs.
[INDIRECT_PROMPT_INJECTION]: The skill provides capabilities to read files and directory listings from remote sandboxes, creating a surface where untrusted data could be ingested into the agent's context. 1. Ingestion points: sandbox.fs.read, sandbox.fs.ls in SKILL.md. 2. Boundary markers: None explicitly documented for content isolation. 3. Capability inventory: File writing, command execution, and network preview creation. 4. Sanitization: No explicit content sanitization or validation routines are described for external data.
[SAFE]: All external downloads, remote repositories, and package references (e.g., GitHub, GHCR, NPM, PyPI) belong to the verified skill author or trusted services. Authentication practices using environment variables or configuration files follow industry standards for developer tools.

blaxel-sdk