The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on local execution of the starwork CLI for project diagnostics (doctor) and updates (upgrade). These commands are gated by strict instructions requiring user confirmation and the use of --dry-run flags before actual modifications occur.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external project files (such as README.md, AGENTS.md, and system rules) and CLI outputs to formulate upgrade plans.
Ingestion points: Files to be read are identified in references/diagnosis-flow.md and references/rules-extraction-guide.md.
Boundary markers: The instructions explicitly distinguish between "facts" provided by tools and the agent's own "inferences" to prevent misleading the user.
Capability inventory: Subprocess calls for starwork doctor and starwork upgrade are documented in the diagnostic and upgrade flow references.
Sanitization: The skill implements human review checkpoints and an explicit "dry-run" requirement before any file-write operations are authorized.
[DYNAMIC_EXECUTION]: The skill generates JSON-formatted "blueprints" and Markdown rule snippets based on its analysis of the local workspace. These generated files are intended for consumption by the StarWork CLI rather than direct execution by the shell, which minimizes the risk of arbitrary code execution through generated content.

starworkDoctor