Plan Mode
Plan Mode is for work that should be designed before anything changes. The assistant researches your workspace, asks clarifying questions, writes a durable plan file, and waits for your approval—then hands off to Agent mode to implement.
Best for lawyers and case teams: multi-step case organization, evidence workflows, document restructures, or any task where you want a clear written plan before the assistant edits files.
When to Use Plan vs Agent vs Ask
Plan Mode
Plan-first workflow. Research and alignment before any implementation.
Features
- • Read and search your workspace (and optional web references)
- • Ask clarifying questions before committing to an approach
- • Write a reviewable .plan.md under .safeAppeals/plans/
- • Handoff via Start Implementation — does not implement itself
Example prompt
"Plan how to reorganize the Evidence folder, build a timeline of key dates, and list gaps before we change any files."
Agent Mode
Autonomous multi-step work with tools. Use when you already know what should happen.
Features
- • Full tool access for edits, organization, and multi-step tasks
- • Can execute a plan after you approve it
- • Best when scope is clear and you want action, not design
Example prompt
"Execute the approved evidence reorganization plan and update the timeline."
Ask Mode
Conversational help for exploration, drafting assistance, and quick questions.
Features
- • Natural Q&A and drafting help
- • Safe explore-style research without a formal plan artifact
- • Ideal for short prompts and iterative writing
Example prompt
"Help me draft a clearer opening paragraph for this appeal letter."
The Plan Mode Flow
- Research — The assistant reads and searches your case workspace (files, folders, and related context) to understand what already exists.
- Clarify — It asks questions when assumptions would change the approach, instead of guessing on material decisions.
- CreatePlan — It writes a plan file under
.safeAppeals/plans/*.plan.mdwith YAML frontmatter (name,overview,todos) and a structured markdown body. - reviewPlan — You review the plan in the approval UI, request changes, or approve it.
- Start Implementation — After approval, use Start Implementation (or switch to Agent) so execution happens in Agent mode. Plan Mode itself does not implement.
Typical cycle
Discovery → Alignment → CreatePlan → reviewPlan → (revise if needed) → Start Implementation
The cycle is iterative. If your answers change scope, Plan Mode loops back to research before rewriting the plan.
Where Plans Live
Plans are workspace files under .safeAppeals/plans/, named like {slug}_{hash}.plan.md. They are first-class artifacts you can open, diff, and revisit—not chat-only drafts.
Each plan file includes YAML frontmatter such as:
name— short display titleoverview— one-line summarytodos— checklist items withid,content, andstatus(pending,in_progress,completed,cancelled)isProject— whether the plan is treated as a longer project plan
Legal workspace tip
Plan files can contain case strategy and client-sensitive detail. For legal/client workspaces, add .safeAppeals/plans/ to .gitignore so plans stay local and are not committed or pushed to a remote.
What Plan Mode Will and Won't Do
- Read-only for case files — Plan Mode does not edit, create, or delete your documents except through CreatePlan writing plan files under
.safeAppeals/plans/. - No Copilot memory path — planning uses workspace research and your answers, not a memory-store workflow.
- One concrete approach — plans commit to a single recommended approach (not open-ended Option A / Option B menus).
- Handoff for execution — after approval, Start Implementation switches work into Agent mode, where domain tools (documents, timeline, email drafts, timer, etc.) become available. The agent still cannot send email — only read, organize, and create drafts.
How to Start
- Open the AI chat sidebar (
Ctrl+L) - Select Plan from the mode dropdown
- Describe the goal in plain language (what outcome you want, constraints, and files involved)
- Answer clarifying questions, review the plan, then approve and start implementation
Recommendation: Use Ask for quick questions and drafting help, Plan when the work needs a written approach first, and Agent when you are ready for autonomous multi-step execution.