A cleanup assistant for AI-assisted development workflows that preserves engineering knowledge, removes temporary clutter, and keeps repositories healthy.
Workflow Janitor is an AI coding workflow skill designed to clean up after large development sessions.
AI agents often create temporary artifacts during implementation:
- Investigation notes
- Generated reports
- Debug files
- Evaluation outputs
- Scratch scripts
- Temporary workspaces
- Agent progress files
Some of these files are disposable. Others contain valuable reasoning, decisions, and project context.
Workflow Janitor performs an audit-first cleanup workflow:
- Analyze workspace artifacts
- Identify valuable knowledge
- Preserve important information
- Remove unnecessary clutter safely
- Generate an auditable cleanup report
It does not blindly delete files.
AI-assisted development creates a new repository maintenance problem.
A typical coding session may leave behind:
project/
├── debug-output.txt
├── investigation-notes.md
├── temp-script.py
├── evaluation-results/
├── generated-reports/
└── useful-architecture-notes.md
Traditional cleanup tools cannot understand the difference between:
- temporary noise
- important engineering context
Workflow Janitor helps agents make that distinction.
Workflow Janitor starts in audit mode.
It analyzes:
- Temporary files
- Generated artifacts
- Agent-created reports
- Workspace changes
- Documentation impact
Nothing is removed automatically.
The result is a cleanup proposal:
Cleanup Report
Remove:
✓ debug-output.log
✓ temporary benchmark files
Preserve:
✓ architecture-notes.md
✓ migration decisions
Suggest:
→ Add *.log to .gitignore
Before removing artifacts, Workflow Janitor searches for useful information:
- Architectural decisions
- Implementation discoveries
- Debugging solutions
- Project constraints
- Important workflow outcomes
Useful knowledge can be moved into:
- README files
- Documentation
- Architecture notes
- Project guidelines
Workflow Janitor follows a preserve-first philosophy.
Cleanup rules:
- Never delete without confirmation
- Prefer quarantine over permanent deletion
- Keep changes traceable
- Protect active development work
Cleanup modes:
Default mode.
Actions:
- Scan files
- Classify artifacts
- Suggest cleanup
- Extract knowledge
No modifications are made.
Only runs after explicit confirmation.
Actions:
- Move approved artifacts
- Apply cleanup changes
- Update documentation when needed
- Generate cleanup summary
Development Session
|
v
AI Creates Temporary Artifacts
|
v
Workflow Janitor Audit
|
+----------------+
| |
v v
Preserve Knowledge Remove Noise
| |
+----------------+
|
v
Cleaner Repository
Workflow Janitor supports multiple AI coding environments.
Currently supported:
| Platform | Support |
|---|---|
| Claude Code | Native skill |
| Cursor | Rules integration |
| GitHub Copilot | AGENTS.md |
| Windsurf | Rules integration |
| Cline | Rules integration |
| Aider | Configuration integration |
Install as a Claude Code plugin:
/plugin marketplace add PrantoSMSS/pranto-plugins
/plugin install workflow-janitor@pranto-pluginsThe skill will automatically become available in supported Claude Code workflows.
Run the installer:
python scripts/install.pyThe installer detects available tools and installs the correct configuration files.
python scripts/install.py --allpython scripts/install.py --tool cursorpython scripts/install.py --dry-runpython scripts/install.py --uninstallUse Workflow Janitor after:
- Large feature implementations
- Refactoring sessions
- AI-assisted coding sessions
- Code reviews
- Benchmark runs
- Temporary worktree usage
- Experimental development
Do not run Workflow Janitor during:
- Active debugging
- Incomplete implementations
- Ongoing experiments
- Failed test investigation
- Temporary work that is still needed
Important development knowledge should survive cleanup.
Understanding comes before modification.
Cleanup decisions remain visible and reversible.
Workflow Janitor does not only clean repositories.
It also helps prevent repeated mess by suggesting:
.gitignoreimprovements- Documentation updates
- Better artifact handling
workflow-janitor/
├── SKILL.md
├── .claude-plugin/
│ └── plugin.json
│
├── scripts/
│ └── install.py
│
├── platforms/
│ ├── aider/
│ ├── cline/
│ ├── copilot/
│ ├── cursor/
│ └── windsurf/
│
├── CHANGELOG.md
└── LICENSE
AI coding tools are powerful, but they leave behind traces of their work.
The goal is not to erase those traces.
The goal is to:
Preserve the knowledge. Remove the noise. Keep the workflow clean.
Created by PrantoSMSS