The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands such as 'cat' and 'git' to retrieve configuration and manage version control. It also executes local Python scripts (setup_experiment.py, run_experiment.py) provided within the skill's own script directory to perform experiment setup and evaluation.
[INDIRECT_PROMPT_INJECTION]: The skill establishes a feedback loop by reading historical results (results.tsv) and strategy documents (program.md) to inform its next experimental actions. Ingestion points: Reads historical context and experiment strategies from the '.autoresearch' directory (SKILL.md). Boundary markers: Absent; there are no specified delimiters to isolate historical data from the agent's decision-making instructions. Capability inventory: File writing, git repository manipulation, and shell execution (SKILL.md). Sanitization: None; the agent is directed to directly review and act upon the contents of the history and strategy files to adapt its behavior.

run