focus

name	focus
description	Start (or resume) a Rill workspace from a theme, journal file, or workspace id. Automatically collects related vault context and creates a workspace/{id}/ directory for an interactive Deep Think session. Use when the user wants to begin or continue a workspace-scoped exploration.
gui	{"label":"/focus","hint":"Start a focus session from this file","match":["**/*.md"],"arg":"path","order":10,"mode":"live"}

/focus — Workspace Start / Resume

Conduct all conversation with the user — and write all generated output — in the language defined by .claude/rules/personal-language.md (or the user's input language if that file is absent). Follow the language rules in full — exceptions and translation quality are defined in the Language Rules of .claude/rules/rill-core.md and the vault's personal-*.md overrides, never restated per skill. The English instructions below are for skill clarity, not for output style.

Tool references in this skill (Read, Edit, Grep, Glob, AskUserQuestion, shell) describe intent, not Claude-specific tool calls. Each harness should map them to its native equivalent — Claude Code uses its built-in tools as named (Read / Edit / Grep / Glob / AskUserQuestion / Bash); Codex CLI uses apply_patch / --search / shell / its own question primitive as appropriate.

Starts (or resumes) a workspace from a theme or file. Automatically collects related information and builds a workspace in workspace/, then begins an interactive session.

Arguments

$ARGUMENTS — one of the following:

• Theme text (e.g., voice input optimization) → Start a new workspace
• Journal file path (e.g., inbox/journal/2026-02-13-1950.md) → Start new workspace from that journal
• workspace/ path or id (e.g., workspace/2026-02-13-rill-development/ or rill) → Resume existing workspace
• Omitted → Propose resuming active workspaces if any, otherwise ask for theme

Not accepted: task file paths (tasks/{slug}/_task.md, or the legacy tasks/{slug}.md). ADR-077 retired task-originated /focus. If the user passes a task path, return an error immediately and suggest /solve {slug} instead — tasks are executed in place now, not through a workspace (ADR-076).

Procedure

Phase 0: Workspace Identification

1. If argument is a workspace/ path or an existing workspace id:
   • Read the metadata file in that directory (priority: _workspace.md > _session.md > _project.md)
   • status: active → Resume workspace → Go to Phase 3
   • status: completed → Ask "This workspace is completed. Would you like to reopen it?" via the harness's question primitive. Yes → Set status back to active, go to Phase 3
   • status: on-hold → Ask "This workspace is on hold. Would you like to resume it?" via the harness's question primitive
2. If argument is omitted:
   • Scan all directories directly under workspace/ (exclude daily)
   • Search for _workspace.md OR _session.md OR _project.md with status: active
   • If active workspaces found, display the list and ask the user whether to resume or create new
   • If no active workspaces, ask "What would you like to think about?"
3. If argument starts with tasks/ (either new tasks/{slug}/_task.md or legacy tasks/{slug}.md):
   • Exit immediately with: "Task-originated /focus was retired by ADR-077. Tasks execute in place now — run /solve {slug} to work on this task, which writes artifacts directly under tasks/{slug}/. If you want a standalone Deep Think surface, start /focus <theme> with a theme (not a task path)."
   • Do not create a workspace, do not Read the task file
4. If argument is text or a journal path:
   • Search for related existing workspaces (match on directory name, tags, _workspace.md body)
   • If related workspace found, ask "An existing workspace exists. Resume or create new?"
   • If creating new → Phase 1

Phase 1: Context Collection

Automatically collect information related to the theme:

1. Bulk search: Extract 2-3 keywords from the theme and run a single cross-cutting Grep:

   Grep(pattern="{keyword}", glob="{knowledge,inbox,workspace,reports,tasks}/**/*.md",
        output_mode="files_with_matches", head_limit=30)
   

   • pages/ is excluded from search targets
   • If results contain both inbox/{type}/ and inbox/{type}/_organized/ with same filename, use only the _organized/ version
2. Supplementary search: Only if bulk search didn't find sufficient related files, run additional Grep on individual directories
3. Organize collection results for inclusion in Phase 2's _workspace.md

Phase 2: Workspace Construction

1. Create directory:
   • workspace/{YYYY-MM-DD}-{topic-name}/ (today's date + kebab-case topic name)
2. Generate _workspace.md via rill mkfile (LLMs must never write created by hand — ADR-060):

   rill mkfile workspace --slug {YYYY-MM-DD}-{topic-name} --type workspace
   

   This creates workspace/{YYYY-MM-DD}-{topic-name}/_workspace.md with created (ISO 8601, auto-assigned), type: workspace, id, and status: active already populated. Then Edit the file to fill in the remaining frontmatter fields (name, origin, tags) and the body:

---
created: {auto-assigned by rill mkfile — do not edit}
type: workspace
id: {YYYY-MM-DD}-{topic-name}
name: {Topic title}
status: active
origin: (path of origin journal. Omit for theme-specified starts)
tags: [related-tags]
---

# {Topic Title}

(Theme overview. Why think about this, context)

## Issues to Consider
- [ ] Issue 1
- [ ] Issue 2
- [ ] Issue 3

## Related Files (MOC)
- [Filename](path) — Brief description
- [Filename](path) — Brief description

## Session History
- {YYYY-MM-DD}: Session started. (overview)

## Next Steps
- [ ] First action

3. The first deliverable file (001-{description}.md) is generated naturally during the conversation (not auto-created at session start)

Phase 3: Interactive Phase

After session start (or resume), interact with the following flow:

1. Display the metadata file (_workspace.md / _session.md / _project.md) contents to user (overview, issues, next steps)
2. On resume, collect additional context:
   • Read deliverables within the workspace
   • Related information from knowledge/notes/ based on tags
   • Recent relevant entries from inbox/journal/
   • Related reports from reports/ (newsletters, dailies, etc.) (ADR-061)
3. Proceed with interactive deep-dive with user
4. File-first principle for deliverables:
   • Use rill mkfile for file creation (ensures timestamp accuracy):

     rill mkfile workspace/{workspace-id} --slug {description} --type {type}
     

     Then use Edit to append body to the output path (frontmatter is pre-generated)
   • type: progress | research | analysis | decision | review
   • Write to file (if any of the following apply):
     • Analysis, research, investigation results
     • Reports, surveys, comparison tables
     • Decisions and their rationale
     • Frameworks, design proposals, recommendations
     • Structured output of 3+ paragraphs
   • Keep as text output (don't create a file):
     • Confirmations, questions, short suggestions (1-2 paragraph conversational exchanges)
     • Direction discussions or brainstorming in progress
     • Summary previews before file save (when getting user confirmation)
   • When in doubt, write to file. The workspace's purpose is to leave traces of thinking for knowledge reuse
5. Update _workspace.md (or _session.md) at each conversation milestone:
   • Add new deliverables to "Related Files (MOC)"
   • Update checkboxes for completed issues
   • Append progress to "Session History"
   • Update "Next Steps"
   • Do NOT change status — never transition the workspace to completed from /focus. See the Rules section below for the completion protocol

Deliverable File Frontmatter

---
created: 2026-02-16T17:30+09:00
topic: {topic-name}
type: research
---

Rules

• Never transition status to completed in /focus. Workspace completion is exclusively performed by /close, which runs the mandatory knowledge distillation via sub-agents (ADR-073). Setting status: completed from /focus bypasses distillation and silently loses the session's knowledge. If the user signals that the session is done, or you judge the workspace has reached its completion conditions, propose running /close (e.g., "This workspace looks ready to complete. Shall I hand off to /close to run distillation?") rather than editing the status directly. The only status transitions /focus may perform are the reopen/resume paths already specified in Phase 0 (completed → active on reopen, on-hold → active on resume)
• File-first principle: Save analytical/report-type output to files. Leave them as workspace files, not assistant text output. The workspace's value lies in accumulated deliverables
• Never modify inbox/journal/ and inbox/*/ original files (read-only)
• Workspace files use numbered NNN-description.md naming
• _workspace.md (or _session.md) is continuously updated as the session progresses
• Once created, deliverable files are generally not modified (additions/corrections go in new files)
• When active workspaces exist, prioritize proposing resume
• Include frontmatter in all files
• Prefer _organized/ version when same-named file exists
• Note metadata correction (ADR-046 D46-7): When reading knowledge/notes/ files, handle in two modes: Mode A — Direct fix (no AI judgment needed, 1-2 Edits per fix, < 100ms):
  1. If frontmatter tags contain deprecated tags (tags in taxonomy.md's "Deprecated Tags" table), remove via Edit
  2. If frontmatter tags contain entity IDs (values matching filenames in knowledge/{people,orgs,projects}/), remove from tags and add to mentions in typed format ({type}/{id}) (skip if already in mentions. ADR-053) Mode B — Append to .refresh-queue (no AI judgment, detection only. < 10ms): Append file paths matching any of the following to knowledge/.refresh-queue (only if not already present):
  • tags is empty array []
  • tags has only 1 tag and that tag has 50+ uses (generic tag only)
  • mentions field does not exist
  • related field does not exist
  • type is not one of record / insight / reference ※ Mode A + Mode B combined have virtually no impact on main task accuracy/speed. Mode B targets are refreshed in the next /distill Phase 0.5
• Backward compatibility: Properly read workspaces that only have _session.md or _project.md. Do not rename during Phase 3 interaction
• Body link rule: When referencing files in workspace deliverables or _workspace.md, use [display name](relative-path) Markdown links. Backtick-only ID references are prohibited

Compaction Resilience

Long /focus sessions may trigger automatic context compression (compaction). Handle with:

• Immediate persistence: Don't leave analytical output in conversation text — immediately write to workspace files (File-first principle reinforcement)
• Regular _workspace.md updates: Update session history and "Next Steps" every 2-3 conversation exchanges
• Context loss recovery: If previous conversation content feels vague, re-read _workspace.md and latest deliverables before continuing