The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various starwork CLI subcommands to maintain project state. These actions are governed by strict safety rules, including the use of --dry-run and requiring explicit user confirmation before executing any permanent write operations.
[PROMPT_INJECTION]: While the skill ingests project-specific context from local Markdown files (e.g., AGENTS.md), it mitigates indirect injection risks through structured boundary markers in its message rendering and by validating the current session ID before performing cross-agent actions.
[SAFE]: The skill's design prioritizes user oversight, explicitly directing the agent to downgrade to manual handoffs if automated tools are unavailable or if session identity cannot be verified.

starworkMultiagent