The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches the uv package manager installer from Astral's official domain and clones the research repository from GitHub.
[COMMAND_EXECUTION]: Executes shell scripts for setup, hardware checks, and loop management, including the execution of the uv installer script and training commands via the command line.
[REMOTE_CODE_EXECUTION]: Implements an autonomous loop where the agent modifies Python code in train.py and executes it to perform machine learning experiments.
[PROMPT_INJECTION]: The skill processes directives from a human-authored program.md file, creating a surface for indirect prompt injection.
Ingestion points: program.md (read by the agent as its research charter).
Boundary markers: Absent; there are no explicit delimiters or instructions to ignore potential commands within the directives.
Capability inventory: The skill supports file writes, Git operations, shell command execution through multiple scripts, and arbitrary Python execution via uv run.
Sanitization: No validation or filtering of the input directives is performed before they are used to guide the agent's actions.

autoresearch