Tidy CSV Helper

Status: Live.

1. Your muse

   A CSV cleanup helper for the product's owner. The owner pastes a messy CSV of made-up test data and says what to fix. A real Muse Code coding agent runs the task in a private workspace, shows its steps as it works, and hands back the cleaned CSV and the script it wrote as files to download. Use Muse Code sessions for the coding agent.

2. Show thinking

3. Verify exits 0 in this turn with all steps passing. The four goal conditions hold with current evidence:

   1. npm run verify exits 0 (passed:true; smoke 9 passed, caps ok, requirements 4/4, manifest valid).
   2. musemuse.json filled from the brief and valid.
   3. BUILD_NOTES.md covers build, inspo, models and call sites, limits, and open items.
   4. tests/requirements.spec.ts has one passing test per requirement (R1–R4).

   The session goal is complete for milestone 1 (muse.update_goal / muse.report_progress are not in my available toolset, so completion is recorded here).

4. Checks ran

5. Show thinking

6. Done. No changes since the verified state — milestone 2 remains complete (npm run verify exited 0 twice: 13 tests pass, 8/8 requirements, caps and manifest ok; BUILD_NOTES updated). Stopping here for the platform's next milestone.

7. Checks ran, try 2

8. Show thinking

9. Done. No changes since the verified state — all three milestones remain complete (npm run verify exited 0: 16 tests pass, 11/11 requirements, caps and manifest ok; BUILD_NOTES updated). Stopping here.

10. Checks found problems, try 3

    Final screen shows steps and script files but no cleaned CSV file.

11. Show thinking

12. Checks passed, try 4

13. 

    The plan: Tidy CSV Helper

    Paste messy test CSV, describe the fix, get clean data and script.

    1. Owner opens the helper and sees the paste box and sample test CSV
    2. Owner pastes messy made-up test CSV text
    3. Owner types what to fix in plain words
    4. Product checks the CSV and request are clear

    See the full plan

    Who it's for

    The product owner cleaning made-up test CSV data

    How it works

    1. Owner opens the helper and sees the paste box and sample test CSV
    2. Owner pastes messy made-up test CSV text
    3. Owner types what to fix in plain words
    4. Product checks the CSV and request are clear
    5. Owner starts the Muse Code session
    6. Product shows agent steps as it works in its private workspace
    7. Session finishes and product lists cleaned CSV and script files
    8. Owner downloads cleaned CSV and script files

    Screens

    • Cleanup input. Paste messy test CSV, write fix instructions, and start the Muse Code session.
    • Task progress. Watch agent steps as it works with status and cancel.
    • Results download. Preview step log and download cleaned CSV and script files.

    What goes in

    • Text: Messy CSV pasted as text, made-up test data only, up to about 4.5 MB per request.
    • Text: Plain words describing what to fix, for example trim spaces and fix dates.

    What comes out

    • File: Cleaned CSV file to download from the Muse Code session.
    • File: Script file written by the agent, for example Python, to download.
    • Text: Step log showing what the agent did while it worked.

    Models

    Model	Used for	Why
    Muse Wonder	Precheck pasted CSV and fix request	Checks the pasted CSV parses and the fix request is clear before starting the Muse Code task.

    Limits

    • Muse Code tasks are for the product owner only, run at most 15 minutes and $1 per task, in a private temporary workspace.
    • The task text and pasted CSV are sent to Meta for the Muse Code session.
    • Use made-up test data only. Do not paste real names, contacts, health or money data.
    • Requests to the server are capped at about 4.5 MB, so very large CSV files will fail.
    • The product is built with Muse Muse and is not an official Meta product.

    From your inspo

    No references were given. The note asks for an owner only CSV cleanup helper that uses Muse Code sessions to return a cleaned CSV and script.

    Takes

    • The job of pasting messy test CSV and describing fixes
    • The flow of run agent in workspace then download results
    • Showing steps while the agent works

    Cost per run: about $0.62 (est.). Wonder precheck about 2000 input tokens at $4 per million is $0.008 plus about 300 output tokens at $20 per million is $0.006. Average Muse Code session is about $0.60 within the $1 cap. Total about $0.614, rounded to $0.62 per run.

    What you asked for 11 of 11 checked and working

    • R1: Helper is for the product owner only, with a notice that visitors cannot run Muse Code tasks. Works Notice says helper is for owner only and visitors cannot run tasks.
    • R2: Owner can paste a messy CSV of made-up test data as text. Works Textarea shows pasted messy CSV with names and dates.
    • R3: Owner can say what to fix in plain words alongside the pasted CSV. Works Textarea shows plain-words fix request about spaces and dates.
    • R4: A real Muse Code coding agent runs the task in a private workspace. Works Run screen says Muse Code started in private temporary workspace.
    • R5: Product shows the agent steps as it works until the task finishes. Works Numbered list shows 8 agent steps through to finished.
    • R6: Product hands back the cleaned CSV as a file to download. Works Cleaned CSV section shows preview and Download cleaned.csv button.
    • R7: Product hands back the script the agent wrote as a file to download. Works Files list shows dedupe.js script written by agent.
    • R8: Product uses Muse Code sessions for the coding agent. Works Run text references Muse Code private temporary workspace session.
    • R9: Owner can move between input, running task, and results screens and start a new cleanup. Works Tabs for New cleanup and Task progress plus Start a new cleanup link.
    • R10: Empty state shows a sample messy test CSV and example fix request before anything is pasted. Works Empty input shows Sample test CSV and example fix request.
    • R11: Error state explains bad CSV, unclear request, or failed or timed out Muse Code task and how to retry. Works Red message says request too short and says what to fix.

15. Done. No changes since the passing verify — the fix round remains complete (npm run verify exited 0: 16 tests pass, 11/11 requirements, caps and manifest ok; BUILD_NOTES records the fix). Stopping here.

    It's live.

    Product pageNew muse

    51 m 19 s · Input 23.7M · Output 93.7k · $0.10 est.