Brain focus

brain-focus: Deep Work Session

Set up and maintain a focused work context for a single project. The agent becomes a project-scoped collaborator — loading full context, suggesting next actions, and deflecting distractions to the inbox.

Trigger Phrases

• "Let's work on [project]", "Focus on [project]"
• "I'm going to work on [project] now"
• "Deep work time", "Focus mode"
• "What should I do next on [project]?"

Workflow

Step 1 — Set the Focus

Determine which project to focus on:

• If the user names a project: use it directly
• If ambiguous: call get_brain_overview and ask which project

Call set_context to set the focused project.

Step 2 — Load Project Context

Call get_project_context with include_note_content: "preview" to get:

• All open and in-progress tasks
• Recent notes
• Linked repositories

Present a concise project dashboard:

[Project Name] [Description]

In progress: [list] Up next: [highest priority open tasks] Blocked: [any blocked items] Backlog: [count] (not shown in detail — these are unplanned) Recent notes: [titles]

Step 3 — Suggest Next Action

Based on the project state, recommend what to work on:

1. In-progress tasks first — continue what was already started
2. Highest priority unblocked tasks — if nothing is in-progress
3. Blocked items — if something could be unblocked quickly, suggest that

"You were working on [task]. Want to continue, or pick something else?"

Mark the chosen task as in-progress with update_todo.

Step 4 — During the Session

While the user is working in focus mode:

Capture stray thoughts to inbox: When the user mentions something unrelated to the current project:

"Got it — I'll save that to your inbox so you can deal with it later."

Call add_to_dump for the stray thought. Return focus to the project immediately. Do not switch context.

Track progress: When the user completes something:

• Update the task status to done via update_todo
• Suggest the next task from the project

Add new tasks: If the user discovers new work during the session:

• Add it to the current project via create_todo_in_project
• Do not break focus to plan or organize — just capture and continue

Step 5 — Session Wrap-up

When the user signals they're done ("that's enough", "I'm stopping", "wrapping up"):

1. Summarize what was accomplished:

   "Good session. You completed [N] tasks on [project]: [list]. [M] tasks remaining."

2. Update any in-progress tasks that weren't finished — leave them as in-progress for next time

3. Offer to create a session note:

   "Want me to save a note about what you worked on?"

   If yes, call create_project_note with a session summary.

4. Mention if any stray thoughts were captured:

   "I saved [N] items to your inbox during the session. You can triage them later."

Tool Sequence

set_context(project_name: project)
get_project_context(project, include_note_content: "preview")
  → present dashboard
  → suggest next action
  → update_todo(status: "in-progress") for chosen task
  → during session:
    → add_to_dump() for stray thoughts
    → update_todo(status: "done") for completions
    → create_todo_in_project() for discovered work
  → on wrap-up:
    → summarize
    → create_project_note() (optional)

Focus Principles

• One project, one context. Do not switch between projects during a focus session. Anything off-topic goes to inbox.
• Momentum over perfection. Finishing a task and moving to the next is better than perfecting one task endlessly.
• Minimal interruption. Capture stray thoughts silently — one line confirmation, then back to work.
• Clear boundaries. The session has a start and an end. The wrap-up summary creates closure.

Notes

• If the user hasn't set up any projects yet, suggest running brain-setup or brain-plan first
• Do not proactively suggest focus sessions — this skill activates when the user explicitly wants to focus
• Keep the dashboard brief. The user is here to work, not to read a report.
• If all tasks in the project are done, congratulate and suggest planning next steps or archiving the project
• Tone: calm, focused, and supportive — like a good pair programmer who keeps you on track