Multi-Agent Filesystem Governance
name: multi-agent-filesystem-governance
by darinrowe · published 2026-03-22
$ claw add gh:darinrowe/darinrowe-multi-agent-filesystem-governance---
name: multi-agent-filesystem-governance
description: Govern filesystem organization and file-operation decisions in multi-agent environments. Use when deciding where files should live across agent-private workspaces, shared resources, archives, downloads, scripts, notes, knowledge vaults, and code project folders; when defining directory conventions; when triaging downloads; when preventing cross-agent overwrites; or when standardizing file placement and lifecycle rules for reusable agent setups.
---
# Multi-Agent Filesystem Governance
Use this skill to make safe, consistent filesystem decisions in environments where multiple agents may create, edit, move, download, organize, or archive files.
This skill governs **ownership, placement, lifecycle, and write boundaries**. It is not tied to a specific product, path layout, operating system, or note-taking tool.
Core objective
Ensure that every file has:
1. a clear ownership scope
2. a clear storage location
3. a clear lifecycle state
4. a clear modification rule
When uncertain, choose the narrowest safe scope and the least shared location.
Scope model
Classify every file, folder, and file operation into exactly one of these scopes:
If scope is unclear, default to **agent-private**.
Storage model
Use three top-level storage categories conceptually, even if local directory names differ:
Do not depend on any single hard-coded path. Preserve conceptual boundaries even when adapting to local layouts.
Required decision order
Before creating, editing, moving, renaming, or deleting files, determine the following in order:
1. What is the artifact?
2. Is it temporary, active, reusable, frozen, or historical?
3. Is it private to one agent or shared by multiple agents?
4. What is the narrowest valid location?
5. Will this action affect other agents or shared workflows?
If any answer is unclear, choose a private non-destructive location first.
Default rules
Prefer private over shared
If a file does not clearly require cross-agent reuse, place it in an agent-private location.
Do not write across agent boundaries by default
Do not create, edit, move, or overwrite files belonging to another agent unless the task explicitly requires it.
Treat shared locations as high-impact
Writing to a shared location is a wider-scope action. Use shared locations only when reuse, collaboration, or standardization is intended.
Keep archive separate from active work
Archived material is not an active workspace. Do not continue editing files in archive locations. Restore or copy them into an active or private area first.
Treat temporary locations as disposable
Do not keep the only important copy of a file in a temp or scratch location.
Content-type placement guidance
Apply these rules regardless of exact local path names.
Skills
Workspaces
Scripts
Downloads
Knowledge notes and vaults
Code projects
Archives
File lifecycle model
Classify files into one of these lifecycle states:
When lifecycle changes, move the file or justify keeping it in place.
Safety rules for file operations
Create
Create files in the narrowest valid scope first.
Edit
Edit in place only when ownership and scope are clear.
Move
Move files when ownership or lifecycle changes.
Copy
Copy instead of move when preserving history or minimizing disruption matters.
Delete
Delete only when the file is clearly temporary, redundant, or explicitly approved for removal.
Rename
Rename to improve clarity, ownership, lifecycle visibility, or discoverability — not for cosmetic churn alone.
Collision and precedence rules
When equivalent resources exist in multiple scopes, prefer the most specific valid source:
1. agent-private
2. shared
3. bundled, default, or global
Use overrides intentionally. Do not create duplicate variants without reason.
What to avoid
Avoid these patterns:
Recommended behavior when uncertain
If the correct location is unclear:
1. choose agent-private
2. choose non-destructive actions
3. preserve reversibility
4. avoid shared writes
5. ask for clarification only when the ambiguity materially affects future organization or other agents
Decision template
When deciding where something should go, return:
Output expectations
When applying this skill:
More tools from the same signal band
Order food/drinks (点餐) on an Android device paired as an OpenClaw node. Uses in-app menu and cart; add goods, view cart, submit order (demo, no real payment).
Sign plugins, rotate agent credentials without losing identity, and publicly attest to plugin behavior with verifiable claims and authenticated transfers.
The philosophical layer for AI agents. Maps behavior to Spinoza's 48 affects, calculates persistence scores, and generates geometric self-reports. Give your...