Manage Project

Manage one project through owned, evidence-backed tasks; this skill coordinates the work rather than replacing each task's specialist workflow.

Use this global workflow-model:

• Run every step in a new, fresh subagent (by whatever means are available to you).
• Keep the invoking agent as the single accountable project owner; subagents may plan, execute, monitor, or review, but the owner maintains the project record, resolves handoffs, and reports the result.
• Give every step, workflow input, and output a unique kebab-case identifier and bind each input or output ID to $<id> in task text.
• Treat only Wait inputs as dependencies; every Use value is available initially.
• Launch all ready steps concurrently and launch newly ready steps immediately.
• A step is ready only when its inputs, context, tools, permissions, and environment are available.
• When a requirement is unavailable, mark the affected work Blocked, record the unmet condition and its review event or time, and do not launch dependents.
• After each meaningful action, persist artifacts, evidence, decisions, status changes, and the next action through the CLI into the project record.
• Start asynchronous work once, save its handle and output location, and inspect it only when its completion event occurs.
• Keep task status equal to its current condition: Not started, Ready, In progress, Blocked, In review, or Done.
• Move no task to Ready without the complete Ready gate and move no task to Done without review of both its definition of done and outcome.
• Use references/ai-agent-processes.md as the global task-management-reference for field meanings, gates, transitions, and role responsibilities.