Create Project

When an AI agent is working on a multi-step implementation task, the conversation context grows large and becomes costly. This skill helps the agent offload the plan, task list, and research findings into durable project files so that:

• The agent can resume work efficiently by reading these files instead of re-deriving the plan from conversation history.
• The user has a persistent record of decisions, architecture, and pending tasks.
• Future agents (or the same agent in a new session) can pick up where the previous one left off.

Workflow

When the user asks you to plan, design, or organize research for a project, follow this workflow:

1. Check existing docs — Read docs/PLAN.md, docs/TODO.md, and any files in docs/references/ if they already exist. Understand the current state before overwriting.
2. Decide what to capture — Ask yourself: what does the next agent (or future me) need to know to continue without re-asking the user?
3. Resolve ambiguities before writing — If the user's request leaves important design choices undefined (e.g., which auth provider, which database), ask the user to decide before writing PLAN.md. PLAN.md must contain zero undecided items. Do not proceed to step 4 until every ambiguity is resolved.
4. Write docs — Create or update the files described below.
5. Summarize in conversation — After writing, give the user a brief summary of what was saved and where.

docs/PLAN.md