The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute arbitrary shell commands defined in a local configuration file (loop.run.yaml), specifically through the gate_cmd and metric_cmd parameters.
[COMMAND_EXECUTION]: The instructions explicitly direct the agent to 'not pause for permission' once the loop begins, enabling the autonomous execution of multiple shell commands and file modifications (up to the defined budget) without human oversight.
[INDIRECT_PROMPT_INJECTION]: The skill presents an indirect prompt injection surface because it reads and modifies files specified in <editable_files> and then executes commands against them. Malicious instructions embedded in the code or query being optimized could attempt to subvert the agent's logic during the 'Apply change' phase.
Ingestion points: Source code or SQL files defined in the editable_files configuration parameter.
Boundary markers: Absent. The skill does not use delimiters or instructions to ignore embedded content when reading the target files.
Capability inventory: The skill performs file system writes (snapshots and edits) and shell command execution via user-defined gate and metric commands.
Sanitization: Absent. No validation or filtering is applied to the content of the files being optimized before the agent processes them.

optimize-loop