The Agent Skills Directory

[SAFE]: The skill enforces strong process isolation by requiring the agent to use platform-specific worker mechanisms, such as isolation: "worktree", ensuring that sub-tasks are performed in isolated environments that do not have direct, unreviewed access to the primary codebase.
[SAFE]: It implements defensive patterns against indirect prompt injection by instructing the orchestrator to treat all worker reports as fallible, requiring independent re-computation of verification metrics and manual inspection of all diffs before acceptance.
[COMMAND_EXECUTION]: The instructions include standard development workflow commands using git and npm for merging work from isolated branches and running project synchronization scripts. These commands are executed locally and are consistent with standard software development practices.

orchestrate-mode