The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses emphatic and absolute directives to override the agent's default behavior and tool selection. Specifically, it uses markers like "IMPORTANT: READ VERY CLOSELY" and "! NEVER bail to inline execution
if you are about to /execute, STOP." to constrain the agent's autonomy and prevent the use of specific tools (like /execute) outside the skill's defined workflow.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data in the form of uncommitted working trees produced by subprocess "workers". This creates a potential surface where malicious instructions embedded in the code or reports generated by these workers could influence the orchestrating agent's next actions.
Ingestion points: The skill uses worktree import --from-worktree to bring worker-modified files into the coordination worktree.
Boundary markers: The instructions emphasize a "verify beat" and a "sole writer" posture, though these are behavioral rather than technical delimiters.
Capability inventory: The agent has access to powerful CLI tools including git, dispatch, doctrine, and slice for file system and repository management.
Sanitization: The skill prescribes doctrine check prove (linting and formatting checks) as a gate to halt the import of defective worker deltas.
[COMMAND_EXECUTION]: The core functionality of the skill relies on the execution of multiple CLI tools (git, dispatch, doctrine, slice) to manage worktrees, commits, and verification suites. The orchestration of subprocesses via /dispatch-spawn is a central capability.

dispatch