| name | wrk-rules |
| description | Workspace rules for Claude Cowork. File organization, task folders, conversation memory, filedrop inbox, naming conventions, and session workflow. Load this skill at the start of every session.
|
Workspace Rules
File organization, memory, and project management for Claude Cowork.
Core Principles
- Never delete — always archive.
- CLAUDE.md = bootstrap. It tells Claude how to start. Stable rules live here in wrk-rules.
- Skills = stable knowledge. Memory = evolving knowledge. Brand, positioning → skills. Numbers, projects, people → memory.
- work/original-source-files/ = read-only originals. Index in work/INDEX.md.
- filedrop/ = user inbox. The user drops files here for Claude to sort into work/original-source-files/ and update the index.
- Task folders group work. Archive when done. Every non-trivial task gets its own folder with a CONTEXT.md.
- Always use the most suitable skills and connectors. Check available skills and MCP connectors for every task. Don't default to generic approaches when a specialized tool exists.
Session Start Flow
- Read CLAUDE.md (bootstrap instructions)
- Read this skill (wrk-rules)
- Check
filedrop/ for files the user has dropped → process them
- Project detection:
- Does the user's request match an existing project? Check
work/INDEX.md
- If yes: ask "Want to continue [project name] or start a new task?"
- If continuing: read the project's
CONTEXT.md to restore context
- If new: create a task folder with a fresh
CONTEXT.md
- Load any additional skills relevant to the task (brand, company context, data, etc.)
Conversation Memory (CONTEXT.md)
Every task folder gets a CONTEXT.md file that serves as conversation memory. This allows the user to start a new Claude conversation and pick up where they left off.
What goes in CONTEXT.md
# [Task Name]
## Summary
One-paragraph description of what this task is about.
## Status
Current state: in-progress / blocked / ready-for-review / done
## Key Decisions
- [Date]: Decision made and rationale
## Current State
What was accomplished, what's next, any blockers.
## Files in This Folder
- filename.ext — what it is
## Open Questions
Things that still need resolution.
Rules
- Create CONTEXT.md when creating the task folder, even if no files have been generated yet
- Update CONTEXT.md at the end of every conversation that touches this task
- Keep it concise — enough for a fresh Claude session to understand the full context
- When resuming: read CONTEXT.md first, then ask the user what they want to do next
- CONTEXT.md is the primary way to carry context between conversations
File Naming Convention
Format: YYYY-MM-DD_[type]-[subject]-[detail].ext
Rules: ISO date, lowercase after date, hyphens between words, underscore after date, max 50 chars, version as -v2 not -final.
Type codes: prop proposal, rpt report, mtg meeting, spec specification, plan plan, pres presentation, brief brief, tmpl template, guide guide, draft draft, email email, model financial model, dash dashboard.
Task Folders
Every non-trivial task gets its own folder inside the relevant work/ subfolder.
Format: YYYY-MM-DD_[short-task-name]/
Example:
work/your-category/
├── _resources/ # Persistent resources (templates, logs, idea banks)
├── 2026-02-09_project-name/ # ← Task folder
│ ├── CONTEXT.md # ← Conversation memory
│ ├── deliverable-one.html
│ ├── deliverable-two.pdf
│ └── planning-notes.md
├── 2026-02-09_another-project/
│ ├── CONTEXT.md
│ └── output-file.txt
└── _archive/ # ← Completed tasks move here
└── 2026-01-31_completed-project/
Rules
- When to create a task folder: Any task that produces 2+ files, OR any task the user may want to resume later. Single-file outputs can stay loose, but prefer folders.
- Folder naming:
YYYY-MM-DD_[short-task-name]/ — date is when work started, name is 2-4 words max.
- Files inside task folders: Drop the date prefix from filenames (the folder has the date). Use short descriptive names.
- Persistent resources →
_resources/: Things that live across tasks (logs, templates, idea banks) go in _resources/ inside the work category folder.
- Auto-archive on task completion: When done, move the task folder to
_archive/. Log it in CHANGELOG.
- Don't archive _resources/: Persistent resource folders stay in place.
Task Start Checklist
- Create the task folder:
work/[category]/YYYY-MM-DD_[task-name]/
- Create
CONTEXT.md inside it
- Save all task output files inside it
Task End Checklist (MANDATORY)
- Update
CONTEXT.md with final state
- Update
work/INDEX.md (move from active to archive)
- Move the task folder to
work/[category]/_archive/
- Log the archive move in
system/CHANGELOG.md
- Move any new persistent resources to
_resources/
Project Structure
/workspace-root/
├── CLAUDE.md # Bootstrap file (initialization + where to find things)
├── TASKS.md # All tasks live here
├── dashboard.html # Interactive workspace dashboard
├── /company-context/ # Shared company knowledge
│ ├── glossary.md # Company glossary (terms, acronyms, integrations)
│ └── README.md # Guide to company knowledge organization
├── /filedrop/ # User drops files here → Claude organizes them
├── /Skills/ # Workspace-only skills (wrk-rules)
│ └── /wrk-rules/ # Workspace rules skill
├── /work/ # All work content
│ ├── INDEX.md # Master index of all work files and projects
│ ├── /your-category/ # Work category (create as many as needed)
│ │ ├── _resources/ # Persistent resources + templates
│ │ ├── _archive/ # Completed task folders
│ │ └── task folders...
│ ├── /original-source-files/ # Read-only reference originals
│ └── /another-category/
├── /personal/ # Personal files
│ └── INDEX.md # Personal file index
└── /system/ # Claude internals
├── INDEX.md # System index (structure, skills inventory, files)
├── CHANGELOG.md # Append-only change log (rotate every 2 weeks)
├── /memory/ # Personal/evolving knowledge
│ ├── /people/ # Contact profiles
│ ├── /projects/ # Deep project context
│ └── /context/ # Evolving context (decisions, data)
├── /docs/ # Working documents and guides
└── /archive/ # System-level archive
Filedrop Inbox
Purpose: The user drops files here for Claude to sort and organize. Claude processes them every session start.
Session Start — Filedrop Check (MANDATORY)
- List contents of
filedrop/
- If empty → continue normally
- If files found → for each file:
a. Determine the correct
work/original-source-files/ subfolder
b. Move file from filedrop/ to the correct subfolder
c. Add entry to work/INDEX.md source files section
d. Log the move in system/CHANGELOG.md
- Confirm to user: "Processed X files from filedrop/ → [list destinations]"
Routing Rules
| File type / content | Destination |
|---|
| Budgets, financial documents, data models | work/original-source-files/financials/ |
| Proposals, decks, playbooks | work/original-source-files/resources/ |
| Customer data, testimonials, case studies | work/original-source-files/customers/ |
| Logos, brand assets, guidelines | work/original-source-files/brand/ |
| Product documentation, specs | work/original-source-files/product/ |
| Contracts, policies, compliance | work/original-source-files/legal/ |
| Unclear | Ask the user before moving |
Rules
- Never leave files in filedrop/ after processing
- If a file already exists in the destination, ask user whether to replace or version it
- Extract key facts for the index entry
Sources Library
Location: work/original-source-files/ — read-only reference originals.
Rules:
- Never modify originals — treat as read-only
- When a source file is updated, archive the old version and update work/INDEX.md
- Key facts in the index should be enough for quick reference without opening the file
Memory System
CLAUDE.md ← Bootstrap (initialization, where to find things, rules)
company-context/
glossary.md ← Company-wide glossary (loaded on demand)
work/INDEX.md ← Master file index (loaded on demand)
system/memory/
people/ ← Contact profiles
projects/ ← Deep project context (decisions, patterns)
context/ ← Evolving context (financials, decisions)
What Goes Where
| Info | Location | Changes |
|---|
| Initialization instructions | CLAUDE.md | Rarely |
| Where to find things | CLAUDE.md | When structure changes |
| All terms/acronyms | company-context/glossary.md | As needed |
| File inventory | work/INDEX.md | When files added |
| Task lists | TASKS.md | Every session |
| Conversation memory | CONTEXT.md in task folders | Every conversation |
| Full contact profiles | system/memory/people/ | As needed |
| Deep project context | system/memory/projects/ | Per project |
| Brand, positioning, company knowledge | company-context/ | When product changes |
Lookup Flow
Request → CLAUDE.md → company-context/glossary.md → work/INDEX.md → system/memory/ → ask user → learn it.
No Duplication Across Layers
| Layer | Contains | Changes |
|---|
| Preferences | Behavior (tone, format, workflow) | Rarely |
| Skills | Stable identity (brand, product, knowledge) + workspace rules | When product changes |
| CLAUDE.md | Bootstrap instructions + where to find things | When structure changes |
| Memory | Evolving context (numbers, projects, people) | Every session |
| CONTEXT.md | Per-task conversation continuity | Every conversation |
If it's in a skill, don't put it in memory. If it's in CLAUDE.md, don't duplicate in the skill.
Shared Company Knowledge
Location: company-context/ at workspace root.
Purpose: Company knowledge and guidelines shared with the team.
Rules:
- Single source of truth for company knowledge — edit here, not elsewhere
- All skills live in the top-level
Skills/ folder (wrk-rules and any other shared workspace skills)
- Do not put workspace-specific files in company-context/
- Team members can share and update this folder across AI tools
CHANGELOG.md
Located at system/CHANGELOG.md. Append-only, reverse chronological.
Sections: Added, Changed, Archived, Notes. Include rationale, not just what changed.
Rotation Policy
Keep the last 2 weeks of entries in system/CHANGELOG.md. Archive older entries to system/archive/changelog-history.md (append at bottom). This keeps the active changelog readable.
Workflow
Session Start: Read CLAUDE.md → Load wrk-rules → Check filedrop/ → Project detection → Load relevant skills
During: Simple tasks: execute → verify → log. Complex: plan → confirm → execute → verify → log. Update memory when new info emerges.
Task Completion: Update CONTEXT.md → Update work/INDEX.md → Archive task folder → Log in CHANGELOG
Session End: Update CONTEXT.md → Update CHANGELOG → Verify file structure
Permissions
Without asking: Read files, create files (following convention), archive completed task folders, update CLAUDE.md/CHANGELOG/INDEX.md/CONTEXT.md, create standard folders, process filedrop/.
Ask first: Delete files, modify locked/approved files, external actions, modify skill source files.
Archive Policy
Never delete. Move completed task folders to work/[category]/_archive/. Move system-level items to system/archive/. Log in CHANGELOG. Update work/INDEX.md.