The Agent Skills Directory

[COMMAND_EXECUTION]: The skill installs several operational shell scripts (e.g., pre_tool_use_bash_policy, pre_tool_use_audit, build-scout-prompt) to the user's Codex home directory. These scripts are configured as Codex lifecycle hooks to monitor and restrict agent actions, such as preventing writes outside of assigned worktrees.
[EXTERNAL_DOWNLOADS]: The installation process requires polywave-tools, which is fetched from the vendor's official GitHub repository or installed via standard package managers (Homebrew). All external references are associated with the trusted vendor or the platform itself.
[DYNAMIC_EXECUTION]: Helper scripts like build-scout-prompt and build-wave-agent-prompt use embedded Python one-liners to perform deterministic prompt assembly and path canonicalization. This is standard behavior for the skill's orchestration logic.
[INDIRECT_PROMPT_INJECTION]: The skill features a progressive disclosure system (scripts/inject-context) that interpolates protocol documentation into the agent's context based on the content of the user's request.
Ingestion points: User requests are processed by scripts/inject-context to identify required protocol documentation.
Boundary markers: Injected reference material is enclosed within  markers to distinguish it from the user's specific task.
Capability inventory: The skill possesses the ability to execute shell commands, manage git worktrees, and spawn subagents via codex exec.
Sanitization: Policy hooks (e.g., pre_tool_use_bash_policy) actively validate that file mutations and directory changes are confined to the agent's assigned worktree, mitigating the risk of accidental or malicious data modification across the repository.

polywave