The Agent Skills Directory

[SAFE]: The skill is designed to manage local development tasks and notes. It implements a layered security model using 'Run Contracts' and 'Worker Protocols' to ensure all automated actions are bounded and reviewed.
[COMMAND_EXECUTION]: The skill facilitates task execution through defined modes like 'run-next' and 'run-loop'. It includes detailed instructions for isolating these executions within dedicated git worktrees (e.g., '<worktrees_dir>/shared/') to prevent accidental modification of the primary workspace and requires a review of worker diffs before integration.
[PROMPT_INJECTION]: The instructions do not contain behavior-override patterns or safety-bypass attempts. It defines its own validity checks (e.g., already-done, superseded, blocked) to ensure the agent only implements tasks that are current and legitimate.
[DATA_EXFILTRATION]: No patterns of unauthorized data transmission or network exfiltration were found. The skill explicitly identifies local workspace directories as 'local-only personal state' that must not be staged or committed, maintaining a clear boundary between private execution and shared team knowledge.
[EXTERNAL_DOWNLOADS]: While the documentation addresses the risk of dependency installation and remote synchronization (git pull), it establishes a policy where such actions are restricted based on risk levels and typically require explicit user authorization.

workspace-worklist