The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads and interprets instructions from local configuration files (config.cfg, program.md, results.tsv) at each iteration of the loop. If these files are modified by an external process or another skill, they could influence the agent's behavior during automated execution.
Ingestion points: Files located in .autoresearch/{domain}/{name}/ (config.cfg, program.md, results.tsv).
Boundary markers: None identified; instructions are processed directly.
Capability inventory: Git operations (checkout, add, commit), file writing, and execution of run_experiment.py via Python.
Sanitization: None observed; the agent is instructed to follow strategies defined in the markdown file.
[PERSISTENCE]: The skill implements a persistence mechanism via CronCreate, allowing instructions to execute automatically at specified intervals (e.g., every 10 minutes, daily). While the platform imposes a 3-day limit on these jobs, this represents a recurring execution pattern outside of direct user interaction.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands (git and python) and dynamically edit files at runtime based on the contents of the program.md strategy file.

loop