The Agent Skills Directory

[COMMAND_EXECUTION]: The skill scaffolds a shell script (measure.sh) based on templates and subsequently executes it using bash in both the --run and --status modes. Additionally, it performs autonomous git operations including git add, git commit, and git checkout to manage codebase state during iterations.
[PROMPT_INJECTION]: The skill's autonomous loop relies on instructions from a project-local file (program.md), creating an entry point for indirect prompt injection.
Ingestion points: The agent reads instructions and constraints from scripts/autoresearch/{loop-name}/program.md during the execution cycle.
Boundary markers: No delimiters or safety warnings are implemented to separate instructions from code or to prevent the agent from obeying malicious content within the file.
Capability inventory: The agent has permissions to write to source files, execute shell scripts via bash, and manipulate the repository using git commands.
Sanitization: The skill lacks validation or sanitization logic for the content of the program.md file before it is used to guide autonomous changes.

autoresearch