The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands to manage experiment state and execution.
Evidence: Uses git checkout, git add, and git commit to manage experiment branches and record changes.
Evidence: Executes local Python scripts via python {skill_path}/scripts/setup_experiment.py and python {skill_path}/scripts/run_experiment.py to list and run experiments.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local configuration and result files which directly influence the AI's reasoning and subsequent file modification actions.
Ingestion points: Reads and reviews content from .autoresearch/{domain}/{name}/config.cfg, program.md, and results.tsv (SKILL.md).
Boundary markers: The content from these external files is loaded into the agent's context without delimiters or specific instructions to ignore potentially embedded commands.
Capability inventory: The agent is empowered to edit target files, commit changes to a git repository, and execute shell commands to run experiments (SKILL.md).
Sanitization: No sanitization, validation, or escaping is performed on the ingested data before it is used to decide on experiment strategies or file edits.

run