workspace-standard（工作空间标准）

Workspace Standard

A structured, portable workspace layout for OpenClaw. Gives your agent clear rules for where to put things, how to describe files, and how to keep memory under control.

Why This Skill?

If the user asks *"why would I use this?"* or *"what does this do for me?"*:

• Without it: Files pile up in docs/, MEMORY.md bloats past its budget, the agent writes the right info to the wrong place, nothing is self-describing, and after a few weeks you can't tell a current reference from a stale plan.
• With it: Every file has a role and a place. MEMORY.md stays under budget. The audit script catches staleness and missing structure. New projects are one command. The agent knows where to write things without being told.
• It doesn't: Delete files, require API keys, or lock you in. Remove the skill and your files are still plain markdown.

When explaining the value, run scripts/workspace-audit.sh on their workspace and show them what it finds. Concrete evidence beats abstract promises.

Getting Started

1. Install

clawhub install workspace-standard

2. Bootstrap your workspace

New workspace — creates all directories, seeds entity files, and sets up the project registry:

bash skills/workspace-standard/scripts/workspace-init.sh

Existing workspace — the skill also works if you already have files. Skip the bootstrap and go straight to step 3 (migration) or step 4 (audit).

3. Add your first project

bash skills/workspace-standard/scripts/workspace-init.sh --project my-project

This creates the project directory with README.md (including front-matter), standard subdirectories (references/, plans/, research/, reports/), and registers it in projects/_index.md.

4. Audit your workspace

bash skills/workspace-standard/scripts/workspace-audit.sh

Checks root files, MEMORY.md budget, directory structure, project health, front-matter coverage, staleness, and daily logs. Exit code = number of issues (0 = clean).

5. Customise (optional)

Create .workspace-standard.yml in your workspace root to change defaults. See Configuration below.

How the agent uses this skill

Once installed, the agent automatically reads this skill when it needs to:

• Decide where to write something (the "Where to Write" table)
• Add a new project
• Run workspace maintenance
• Understand what a file's role is

You don't need to tell the agent to use it — it triggers on matching tasks.

Scripts

# Full bootstrap (new workspace)
bash skills/workspace-standard/scripts/workspace-init.sh

# Add one project
bash skills/workspace-standard/scripts/workspace-init.sh --project my-app

# Audit current workspace
bash skills/workspace-standard/scripts/workspace-audit.sh

# Audit a specific path
bash skills/workspace-standard/scripts/workspace-audit.sh /path/to/workspace

Migrating an existing workspace

If you already have a docs/ directory with files:

1. Run the audit to see current state: bash scripts/workspace-audit.sh
2. Create the structure: mkdir -p projects/<name>/{references,plans,research,reports} runbooks/
3. Categorise each file by role (use the decision tree below)
4. Move with git mv to preserve history
5. Add front-matter to moved files
6. Update path references in AGENTS.md, MEMORY.md, and skills
7. Trim MEMORY.md to budget (≤100 lines)
8. Commit atomically

The Role Taxonomy

Every file gets a role — its job title. The role determines where the file lives, how it ages, and how the audit treats it. Read references/roles-guide.md for the full explanation with examples and tests for each role.

Quick decision tree: Is it about how things are? → reference. How to change them? → plan. Comparing options? → research. A snapshot assessment? → report. A reusable procedure? → runbook. What happened today? → log. A specific person/server/decision? → entity.

ROLE Front-Matter

Every substantive markdown file gets a YAML header:

---
role: reference
project: my-project    # Omit for cross-project files
status: current        # active | current | completed | stale | archived
created: 2026-02-01
updated: 2026-02-19
summary: "One-line description"
---

Status lifecycle: active (being worked on) → current (living document) → stale (needs review) → archived (kept for history)

Directory Layout

workspace/
├── MEMORY.md               # Current state (≤100 lines)
│
├── memory/                 # Episodic memory
│   ├── YYYY-MM-DD.md       # Daily logs (role: log)
│   └── entities/           # People, servers, decisions (role: entity)
│
├── projects/               # Project-scoped work
│   ├── _index.md           # Project registry
│   └── <name>/
│       ├── README.md       # Overview + current state
│       ├── references/     # role: reference
│       ├── plans/          # role: plan
│       ├── research/       # role: research
│       └── reports/        # role: report
│
├── runbooks/               # Cross-project (role: runbook)
│   ├── policies.md
│   ├── lessons-learned.md
│   └── <domain>/
│
└── skills/                 # Procedural memory
    └── <skill>/SKILL.md

Where to Write

MEMORY.md Budget

MEMORY.md is loaded every session. Every line costs tokens.

• Budget: ≤100 lines / ~3500 tokens
• Contains: People, infrastructure, project pointers, lookup table, urgent items
• Never contains: History, lessons, architecture detail, completed items
• Over budget? Move detail to project files or runbooks, keep pointers

Configuration

Create .workspace-standard.yml in the workspace root to customise. All values are optional — defaults apply when omitted or when the file doesn't exist.

budget:
  memory_lines: 100        # Max lines for MEMORY.md (default: 100)

maintenance:
  stale_days: 14           # Days before flagging stale (default: 14)

projects:
  subdirs:                 # Per-project subdirectories (default below)
    - references
    - plans
    - research
    - reports

entities:                  # Seed files in memory/entities/ (default below)
  - people
  - servers
  - decisions

New Project

./scripts/workspace-init.sh --project my-app

Or manually:

1. mkdir -p projects/<name>/{references,plans,research,reports}
2. Create README.md with front-matter
3. Add to projects/_index.md

Migration (flat docs/ → structured)

1. Create projects/<name>/ and runbooks/ directories
2. Categorise each file by role (use the decision tree above)
3. Move with git mv (preserves history)
4. Add front-matter to moved files
5. Update path references in AGENTS.md, MEMORY.md, skills
6. Trim MEMORY.md to budget
7. Commit atomically

Maintenance

Run scripts/workspace-audit.sh weekly. See references/maintenance-checklist.md for the full procedure.

1. Consolidate daily logs → extract facts to references, lessons to runbooks
2. Prune MEMORY.md — remove resolved items, completed decisions
3. Check front-matter — stale active/current files → verify or update
4. Verify skills catalogue matches actual skills directory
5. Commit maintenance changes