The Agent Skills Directory

[COMMAND_EXECUTION]: The skill defines procedures for managing isolated worktrees using standard shell commands such as git worktree, mkdir, and grep. These operations are limited to provisioning and cleaning up development environments within the local filesystem.
[REMOTE_CODE_EXECUTION]: The orchestration workflow utilizes a local Python script (skills/worktree-data-sync/scripts/sync_worktree_data.py) to synchronize non-git data between worktrees. This script is part of the internal project infrastructure and does not involve executing code from remote or untrusted sources.
[PROMPT_INJECTION]: The skill is designed to process and act upon feedback from specialized subagents.
Ingestion points: Reviewer findings and implementer status reports are ingested by the orchestrator in SKILL.md during task completion and review cycles.
Boundary markers: The skill uses structured dispatch templates with defined fields (Stage:, Task:, Additionally:) to delineate task parameters and steering instructions.
Capability inventory: The orchestrator has the capability to execute shell commands for worktree management and dispatch new agents via the Agent tool.
Sanitization: SKILL.md includes a mandatory adjudication protocol that instructs the orchestrator to manually verify reviewer findings against the source code to identify and reject false positives before proceeding.

agent-orchestration