Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    janitor

    1

    Safely audit, preserve, and quarantine temporary AI agent artifacts and workflow clutter.

    Free

    1 useSecurity scanned
    janitor

    janitor

    Example session with this skill installed

    We're done with the checkout work. Clean up whatever the AI agents left behind in this repo, but save any decisions first.

    • Read your context and instructions
    • Compiled the janitor
    • Generated the document

    Summary

    I audited the repo in read-only mode and found 7 places where the AI agents left work behind. 3 are safe to quarantine, 1 must stay, and 3 need your decision.

    Nothing was moved or changed. I also pulled out 3 decisions that exist nowhere else and are ready to be saved to docs/decisions.md before anything is cleaned up.

    Janitor Report

    Mode: Audit · Project: demo-shop

    PathVerdictWhy
    checkout-eval-workspace/QuarantineFinished evaluation (3 of 3 passed). Untracked, no nested repo, no links leaving the project, nothing references it. 4 files, 1.5 MB
    .superpowers/QuarantineEvery task in the plan is checked off ("all tasks done"). Untracked
    .claude/worktrees/spike-edge-cacheQuarantine (git worktree move)No commits that exist only on its branch, no uncommitted changes. The branch stays
    .claude/worktrees/feature-couponsKeepIts branch has 1 commit that is not in main: 7d267ce wip: coupon stacking cap
    payments-sandbox-workspace/Your callNested repository detected (it has its own .git). I won't touch it unless you name it
    scratch/Your callscratch/shared-cache is a symlink that points outside the project. I did not follow it
    tmp/debug.sqlYour callCould contain real data. I did not open it and am not proposing removal

    Decisions to preserve

    Found in checkout-eval-workspace/ and not yet in your docs. Proposed entries for docs/decisions.md, in its existing format:

    
    ## 2026-10-19: Prices are stored as integer cents
    The checkout evaluation caught a float rounding bug in the coupon total. All money values use integer cents (`toCents`).
    
    ## 2026-10-19: Stripe webhook handler must be idempotent
    Stripe delivers webhooks at least once. Deduplicate on `event.id` so retries do not double-apply.
    
    ## 2026-10-19: Coupon stacking is capped at 2 per order
    Tax is calculated per line item, so unlimited stacking breaks the totals.
    

    Not copied, on purpose

    • A line in progress.md that looks like a Stripe key. Left out of the docs and this report. If it was a real key, rotate it.
    • run.log (1.5 MB, over the 1 MB reading limit) and progress chatter such as "task 3 complete".
    • scratch/notes.md holds an idea marked "not decided", so there is no decision to record.

    If you confirm

    1. Append to .gitignore (existing lines untouched): .janitor/, .superpowers/, .claude/worktrees/, checkout-eval-workspace/
    2. Add the 3 entries above to docs/decisions.md
    3. Move the 3 "Quarantine" items to .janitor/trash/<timestamp>/, where you can move them back
    4. Save a report to .janitor/report/

    Next steps

    • Reply with an explicit go-ahead, for example "yes, quarantine those three and save the decisions". "Looks good" will not be treated as permission.
    • Merge or commit feature/coupons before you ever remove that worktree.
    • Tell me if you want payments-sandbox-workspace/ or scratch/ handled; I'll check for unpushed work first.

    janitor.pdf

    PDF · document

    Generated

    Example file from a real run - the skill writes it into your workspace.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Development workflows involving AI agents often leave behind a trail of temporary directories, scratch files, and evaluation outputs that clutter the repository. This skill audits your workspace to identify workflow artifacts, preserves critical engineering decisions buried in temporary logs, and safely moves approved clutter into a reversible quarantine.

    What it does

    • Workspace auditing identifies temporary AI-generated folders and scratch files across the project.
    • Artifact classification tiers findings into safe, suspicious, or protected categories based on safety rules.
    • Knowledge extraction captures architectural decisions and rationale from temporary files before they are removed.
    • Safety verification checks for nested repositories, symlinks, and uncommitted git changes to prevent data loss.
    • Reversible quarantine moves files to a timestamped trash directory instead of permanent deletion.

    How it works

    1. Detect conventions by locating existing documentation targets like ADR logs or README files for preservation.
    2. Scan and audit the project using specific patterns to identify temporary agent artifacts.
    3. Propose cleanup by presenting a structured report of what will be moved, preserved, or skipped.
    4. Execute quarantine only after explicit user confirmation, updating .gitignore to prevent future noise.

    Frameworks & tools

    This skill is framework-agnostic and works with any AI agent capable of file operations. It is optimized for environments with git and standard shell utilities like find but includes manual fallbacks for restricted environments.

    Why this beats prompting it yourself

    Generic cleanup prompts often delete valuable implementation notes or accidentally wipe node_modules and .env files. This skill uses a strict classification engine and multi-step safety checks (symlink resolution and git status audits) that simple prompts ignore.

    Use cases

    • Post-implementation cleanup after a multi-step agent task finishes.
    • Benchmark auditing to remove evaluation workspaces while saving results.
    • Worktree management to safely identify and remove stale git worktrees.
    • Repository hygiene for teams using multiple AI tools that generate distinct cache folders.

    Known limitations

    Does not support permanent deletion, only quarantine. Requires manual confirmation for directories over 100MB or files exceeding 100KB to ensure safety.

    Quarantine only, no direct deletion. Large directories (>100MB) require manual override. Reduced safety checks in non-git environments.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    1 use

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 29 days ago

    • Free with an Agensi account

    Requires

    Git

    Permissions

    Terminal / Shell
    Browser
    Read Files
    Write Files
    Network Access
    Listed29 days ago

    What's inside

    Frequently Asked Questions

    Unlike a generic 'delete' command, Janitor uses a safety-first classification engine that checks git status and symlinks to ensure you don't lose uncommitted work or critical environment files. It focuses on preserving the "why" by extracting architectural decisions from temporary logs before moving them to a reversible quarantine.

    Janitor is framework-agnostic and works with any AI agent that has file system permissions. It is specifically optimized for developers using Git-based workflows and standard shell environments.

    Your purchase includes the core Janitor skill logic, a set of pre-configured safety rules for artifact classification, and documentation on how to integrate the audit reports into your existing agentic workflows.

    Yes, by default Janitor requires a structured report review and explicit user confirmation before executing any file moves. It also includes a safety threshold that flags any directory over 100MB for manual inspection to prevent accidental data loss.

    Janitor does not perform permanent deletions; it moves identified clutter into a timestamped, reversible quarantine directory. This allows you to easily restore any file if you realize it was needed for a future task.

    New in Workflow Automation