Multi-Agent Folder Cleanup v1.1.0

Multi-Agent Folder Cleanup v1.1.0

This release makes the skill ready for native ChatGPT and Codex packaging while preserving the original conservative Audit / Plan / Execute safety model.

Downloads

File	Intended host
multi-agent-folder-cleanup-1.1.0-openai.zip	ChatGPT and Codex skills-only plugin
multi-agent-folder-cleanup-1.1.0-portable.zip	Standalone Agent Skill hosts
multi-agent-folder-cleanup-1.1.0-claude.zip	Claude Code / Claude Desktop plugin
SHA256SUMS.txt	Checksums for every archive

Changes

• Added .codex-plugin/plugin.json and agents/openai.yaml for native OpenAI plugin and skill discovery.
• Reduced SKILL.md to a concise routing and safety contract; moved helper command details to references/audit-tools.md while retaining the full protocol in references/workflow.md.
• Fixed three Python/PowerShell parity gaps: versioned release-note claim detection, claim size reporting, and dotfile extension classification.
• Added a deterministic cross-language parity test and pull-request CI.
• Added a separate OpenAI release archive and install guide.
• Synchronized package versions at 1.1.0.

Validation scope

CI validates the skill frontmatter with OpenAI’s pinned validator, checks version and reference consistency, compiles Python, parses PowerShell, and runs both audit helpers against the same fixture. The release workflow also builds, extracts, and exercises all archives.

Not claimed: installation through the public universal plugin directory, which requires a separate OpenAI submission and review; live OneDrive hydration behavior, which requires a representative synced Windows folder; or live end-to-end loading in every supported host.

Multi-Agent Folder Cleanup v1.0.0

Multi-Agent Folder Cleanup v1.0.0

An Agent Skill for auditing, planning, and safely reorganizing a shared project
folder so that any AI agent opening it cold can tell what is true now, what is
proposed, and what is dead - without picking between two copies of the same
document.

Works with Claude, ChatGPT, Grok, Copilot, Codex, Manus, and local models.

Downloads

File	For	Contains
multi-agent-folder-cleanup-1.0.0-claude.zip	Claude Code / Claude Desktop plugin, or Claude.ai skill upload	Plugin layout with .claude-plugin/plugin.json + the skill under skills/
multi-agent-folder-cleanup-1.0.0-portable.zip	ChatGPT, Grok, Codex, local models, Claude.ai skill upload	The bare skill folder + INSTALL.md
SHA256SUMS.txt	Everyone	Checksums for both ZIPs

Each ZIP has its own INSTALL.md with per-host steps.

What's in it

• SKILL.md - the protocol: Audit / Plan / Execute modes, eight-bucket
  classification, the safety rules that stop a half-finished move
• references/workflow.md - full staged-move protocol (D0-D7), record-only
  Execute (D8), report formats, failure recovery
• references/navigation-templates.md - README / STATUS / AUTHORITY / INDEX /
  HANDOFF templates
• scripts/audit_folder.py - cross-platform read-only inventory
• scripts/audit_folder.ps1 - Windows/OneDrive inventory, incl. the cloud
  placeholder check Python cannot do
• scripts/verify_move.py - hash baseline / preflight / verify; never moves
  anything

Design commitments

• Every script is read-only against the target folder.
• Deletion is never in scope by default.
• Execution requires an explicitly approved literal move map - no globs, no
  "and the rest of that folder".
• verify exits nonzero on any mismatch, and a nonzero exit is a stop
  condition, not a warning.

Requirements

• Python 3.8+, standard library only. No dependencies.
• PowerShell 5.1 or 7+ for the .ps1 script (Windows only; optional elsewhere).

What CI checked before publishing

The release workflow ran against this tag and every step passed:

• SKILL.md frontmatter parses; name matches the skill folder; description
  within the length limit; tag matches metadata.version
• plugin.json is valid JSON; Python scripts compile; audit_folder.ps1
  parses under the PowerShell parser
• Both audit scripts ran against the same test tree and both reported the same
  duplicate group
• verify_move.py was run through baseline and verify, and the build fails
  if verify reports success on a move that was never performed
• Both ZIPs were extracted after building and a script was run from inside the
  extracted package

Not checked by CI: OneDrive hydration behaviour, which needs Windows; and
whether a host actually loads the skill, which needs a real install.

Known issue in this release

INSTALL.md inside -claude.zip says to run
/plugin marketplace add JesseRaber/multi-agent-folder-cleanup. That command
fails on this tag: /plugin marketplace add needs
.claude-plugin/marketplace.json, which this tag does not contain. Use the
local-path install, or the -portable package, until a release that includes
the catalog. Fixed on main.

Authority

If a downloaded copy disagrees with this repository, the repository is current.
A ZIP, a chat upload, or a host-local skill cache is not.

MIT licensed.