The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses standard Git commands like git rev-parse and git status to manage the repository state and identify exclusion paths. This is routine behavior for a development-focused agent.
[DATA_EXPOSURE]: The skill defines a local directory structure for storing audit logs, decisions, and temporary artifacts. This promotes transparency and prevents data leakage into the main repository history by updating local Git exclusion files.
[DYNAMIC_EXECUTION]: The instructions allow for the creation of intermediate debugging scripts (Bash and Python) within a dedicated directory. These are intended for workspace-local probes and are kept separate from production build trees.
[PROMPT_INJECTION]: The skill involves analyzing external code and revisions for precedents. While processing external data is a surface for indirect prompt injection, the instructions include safety-oriented constraints such as prioritizing local source over external material and bounding investigations to specific symbols.

zephyr-explore