The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions and a setup script (setup.sh) to download the uv package manager installer from https://astral.sh/uv/install.sh and execute it via a shell pipe. Astral is a well-known service provider, and this is the official installation method for the tool.
[REMOTE_CODE_EXECUTION]: The autonomous research loop functions by having an AI agent edit a local script (train.py) and then execute it using the uv run command. This dynamic execution of agent-modified code is the core purpose of the skill for performing ML research.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes research directives from program.md that influence the agent's code modifications and experimental strategies.
Ingestion points: The agent reads directives from program.md (read-only) and modifies train.py (read/write) based on its interpretation.
Boundary markers: No explicit security delimiters or 'ignore embedded instructions' warnings are present to isolate the directive content from the agent's primary system instructions.
Capability inventory: The agent has access to Bash, Read, Write, Edit, Glob, Grep, and WebFetch tools, and can perform git commit, git reset, and uv run train.py.
Sanitization: The skill does not implement sanitization or validation for the contents of the program.md file before it is processed by the agent.
[COMMAND_EXECUTION]: The skill includes several utility shell scripts (check-hardware.sh, setup.sh, run-loop.sh) that perform system environment checks, repository cloning via git, and ML process execution.

autoresearch