The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends installing the Tensorlake CLI via a piped shell command: curl -fsSL https://tensorlake.ai/install | sh. This is a remote execution pattern; however, it targets the official domain of the skill's author, which is a well-known service in this context.
[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to fetch live documentation from https://docs.tensorlake.ai/llms.txt to use as the primary source of truth. This exposes the agent to potential instructions embedded in external content.
Ingestion points: Instructions in SKILL.md and AGENTS.md direct the agent to fetch and process markdown files from docs.tensorlake.ai.
Boundary markers: Absent. There are no instructions to the agent to disregard or isolate potential prompts found within the fetched documentation.
Capability inventory: The skill documents and enables extensive capabilities including file system operations (write_file, read_file), process management (start_process), and interactive shell sessions (create_pty) as detailed in references/sandbox_sdk.md.
Sanitization: Absent. The skill explicitly directs the agent to trust the live documentation over bundled snapshots.
[DYNAMIC_EXECUTION]: The evaluation tool evals/grade_static.py uses importlib.import_module() to dynamically load the tensorlake package for name verification. This is standard behavior for a static analysis script provided with the skill.
[COMMAND_EXECUTION]: The test scripts evals/run.py and evals/grade.py utilize subprocess.run() to execute agent commands and grading logic. These are standard components of the development and testing harness included in the repository.

tensorlake