The Agent Skills Directory

[SAFE]: The skill establishes a robust, auditable research process using a local Python script for state management and git worktrees for execution isolation. No malicious patterns such as obfuscation, credential harvesting, or unauthorized data exfiltration were identified.
[COMMAND_EXECUTION]: The skill facilitates the execution of user-defined evaluation scripts (dev-eval and test-eval) to assess progress. This is a core functional requirement of the optimization loop and is performed within the user's local development environment.
[EXTERNAL_DOWNLOADS]: Reference documentation is provided for an optional upstream CLI tool from a public research repository (RUC-NLPIR/Arbor). The skill's primary mode of operation is native to the agent environment and does not perform automated external downloads.

arbor