Skip to content

DAAF-Contribution-Community/daaf / .claude/skills/marimo

marimo

Verified

Reactive Python notebook system with cell reactivity, UI elements, SQL cells, plotting, and app deployment. DAAF's standard notebook format — stored as Git-friendly .py files, not .ipynb. For DAAF pipelines, Stage 9 literally archives existing executed scripts in bounded comment/pass cells paired with real execution logs; optional display-only cells may preview existing Parquet data or show already-created figures, but may not analyze data or generate figures. Use when assembling Stage 9 research notebooks, building standalone interactive data apps, or converting Jupyter notebooks to marimo format. R equivalent: quarto — when execution language is R, use quarto instead.

marimo

marimo reactive Python notebook system for reproducible and interactive data work. Covers cell reactivity model, UI elements (sliders, dropdowns, tables, forms), SQL cells, DataFrame display, plotting integration, validation patterns for data pipelines, and deployment as apps, scripts, or WASM. In DAAF Stage 9, its role narrows to literal script-and-log archival, with only bounded existing-data or existing-figure display allowed; the broader interactive capabilities apply outside Stage 9. Use when assembling Stage 9 research notebooks, developing reactive marimo notebooks, building interactive data apps, or converting Jupyter notebooks to marimo's Git-friendly .py format.

Comprehensive skill for building reactive Python notebooks with marimo. Use decision trees below to find the right guidance, then load detailed references.

What is Marimo?

marimo is an open-source reactive Python notebook that automatically keeps code and outputs consistent:

  • Reactive: Run a cell, and dependent cells automatically re-run
  • No hidden state: Delete a cell, its variables are scrubbed from memory
  • Pure Python: Notebooks stored as .py files (Git-friendly)
  • Interactive: UI elements sync with Python without callbacks
  • Deployable: Run as scripts, deploy as web apps, export to WASM

How to Use This Skill

Reference File Structure

Each topic in ./references/ contains focused documentation:

FilePurposeWhen to Read
quickstart.mdInstallation, CLI, marimo check lint rules, first notebookStarting a new project; linting a notebook
reactivity.mdCell execution model, dataflowUnderstanding marimo's reactive model
validation-patterns.mdTransform-validate patterns, checkpointsREQUIRED for data analysis workflows
ui-elements.mdInteractive elements (mo.ui.*)Adding interactivity
sql-data.mdSQL cells, dataframes, plottingWorking with data
outputs-layouts.mdMarkdown, layouts, formattingStyling outputs
apps-deployment.mdApps, scripts, export, deploySharing/deploying notebooks
gotchas.mdCommon errors, best practicesDebugging issues

Reading Order

  1. New to marimo? Start with quickstart.md then reactivity.md
  2. Data analysis workflows? Read validation-patterns.md (REQUIRED for rigorous analysis)
  3. Building features? Read the relevant topic file
  4. Having issues? Check gotchas.md first

The reference-file routing in this skill applies to advisory and brainstorming turns as much as implementation. Recommending an approach, reviewing a plan, or answering a question that touches a routed topic calls for reading the routed reference file just as much as writing code does — the reference files carry curated caveats and environment-specific constraints that this overview and general knowledge lack.

Related Skills

Load these skills together with marimo for comprehensive workflow support:

Always Load Together:

  • data-scientist - Provides validation methodology and EDA principles for the scripts a Stage 9 notebook archives, and documentation patterns for standalone marimo apps (its references/code-documentation.md)
  • polars - DataFrame operations for data transformations within notebooks

Load for Specific Features:

  • plotnine - Static publication-quality plots (ggplot2 style)
  • plotly - Interactive visualizations with hover/zoom

Prerequisite Knowledge: If new to marimo, first understand:

  1. Python basics
  2. DataFrame operations (polars or pandas)
  3. Basic plotting concepts

CRITICAL: Stage 9 is Script COMPILATION, Not Dashboard Building

For data research workflows, the Stage 9 marimo notebook has ONE job: LITERALLY COPY script file contents into cells.

What Stage 9 IS

  • A script viewer — Copy-paste scripts into marimo cells
  • An audit tool — Display execution logs to prove what ran
  • A file compiler — Read files, copy contents, format as notebook

What Stage 9 is NOT

  • ❌ NOT a dashboard builder
  • ❌ NOT an interactive analysis tool
  • ❌ NOT a place for new aggregations or filters
  • ❌ NOT a place for UI widgets (dropdowns, sliders, search)

Stage 9 Notebook Assembly

Stage 9 is handled by the notebook-assembler agent (see .claude/agents/notebook-assembler.md), which:

  1. READS script files from scripts/stage{5,6,7,8}_*/
  2. EMITS a recognizable DAAF Marimo archive identity (import marimo, app = marimo.App(...), and Generated by notebook-assembler agent.)
  3. EMITS one contiguous canonical three-cell bundle per script: a literal mo.md() header whose **Script:** path matches the archive source; an immediately following exact # SOURCE: scripts/<stage>/<filename>.py comment/pass-only archive cell; and an immediately following matching Execution Log (<filename>.py) accordion with one plain-fenced body in a raw mo.md(r'''...''') literal (r""" when the log contains ''')
  4. REQUIRES the accordion body to contain the real, non-empty, non-placeholder execution log copied verbatim; a missing or ambiguous log blocks assembly rather than producing a placeholder
  5. OPTIONALLY ADDS, only after the complete bundle, either a bounded Parquet preview (pl.read_parquet() + head()/mo.ui.table()) or display of an already-created figure (mo.image()); neither path performs analysis or creates/modifies a figure, and display cells are not decompiled

All three path labels match one project-relative scripts/<stage>/<safe-name>.py path, where <stage> is exactly stage5_fetch, stage6_clean, stage7_transform, or stage8_analysis. Absolute paths, backslashes, traversal components, and extra path levels are not canonical.

Archival mechanism: every copied Python line is prefixed with # (a blank line becomes #) and the archive cell ends with exactly one pass # Cell must have executable statement, with no other executable content. This keeps the code visible and searchable without executing it (Marimo cells are live; uncommented copies would collide on imports and variable names). Canonical structural recognition centers on the literal mo.md() script header, preserving the intentional Plan-template mapping; Quarto/R instead centers on its exact VERBATIM COPY marker. Bounded legacy forms accepted by the decompiler are intake compatibility only, not newly emitted output. See .claude/agents/notebook-assembler.md for the emitting template.

ABSOLUTE PROHIBITIONS for Stage 9

The following are NEVER ALLOWED in Stage 9 notebooks:

Prohibited ElementWhy
mo.ui.dropdown()No dropdowns — not a dashboard
mo.ui.slider()No sliders — not a dashboard
mo.ui.multiselect()No multiselects — not a dashboard
mo.ui.text() for searchNo search boxes — not a dashboard
mo.ui.dataframe()No transform editor — it filters, groups, aggregates, and pivots
mo.ui.data_editor()No data editing — display existing data only
.group_by() (new)No new aggregations — copy script code only
.agg() (new)No new aggregations — copy script code only
.pivot() (new)No pivot tables — copy script code only
.filter() in data cellsNo filtering — just load and display
.with_columns() in data cellsNo transforms — just load and display
"Interactive Filters" sectionNot a dashboard
"Data Explorer" sectionNot a dashboard
"Institution Lookup" featureNot a dashboard

The Only Permitted New Display Code

Stage 9 permits exactly two bounded display forms:

  1. Preview existing Parquet data without transforming it:
df_5_1 = pl.read_parquet(PROJECT_DIR / "data/raw/file.parquet")  # one unique name per preview cell
mo.ui.table(df_5_1.head(100), selection=None, show_column_summaries=False)
  1. Display an already-created Stage 8 figure without generating a new one:
mo.image(PROJECT_DIR / "output/figures/existing-figure.png")

No .filter(), .with_columns(), .select(), aggregation, plotting, or figure generation. The preview is bounded to the first 100 rows and shown as a plain read-only table: selection=None drops the row-selection checkboxes (the default is "multi") and show_column_summaries=False drops the per-column summary charts. The image cell only displays an existing artifact.

Both forms read PROJECT_DIR, the absolute project Path the notebook defines in its own cell, and each preview gets its own name. marimo export html resolves a relative path against the shell's working directory, not the notebook's folder; a name assigned in two cells is a breaking MB002 error; and mo.image() given a string path to a missing figure renders a broken image without failing, while a Path fails the cell. So pass the Path itself to mo.image(), never str(PROJECT_DIR / ...).

Tooling Cautions for Stage 9 Notebooks

A Stage 9 notebook is a literal archive that scripts/decompile_notebook.py must be able to read back, so some routine marimo actions damage it:

  • Never marimo check --fix or --unsafe-fixes. Every archive cell is reported as MF004 empty-cells with an unsafe fix that deletes it. Read-only marimo check is fine (see ./references/quickstart.md > Check (Lint)); the Stage 9 check commands live in .claude/agents/notebook-assembler.md.
  • Don't re-save it from the marimo editor. Saving can write cell options into decorators, such as @app.cell(hide_code=True) or, new in 0.25.0, @app.cell(expand_output=True) when an output is expanded. The decompiler accepts only a bare @app.cell line and rejects the notebook.
  • Keep the execution-log accordion keyword-free. mo.accordion(..., expanded=...) (new since 0.19.11) or any other keyword makes decompilation fail.
  • Put each execution log in a raw literal, mo.md(r'''...''') (or r"""...""" when the log contains '''). In a non-raw string Python decodes backslashes in the log: a Windows path C:\new\table silently gains a line break and a tab that marimo check and export html cannot see, and \U or \N make the cell unparsable. The decompiler warns about non-raw log literals that contain backslashes, and that warning fails Step 6c condition (i).
  • Make the log's code fence longer than any backtick run in the log. The fence is three backticks unless the log holds a run of three or more (a printed Markdown fence, for example); then it is one backtick longer than the longest run, and the closing fence has the same length. A log line equal to the fence would close it early and cut the displayed log off; marimo check and marimo export html (Steps 6a and 6b) cannot see that. marimo's pymdownx.superfences closes at exactly the fence length, CommonMark at that length or more, and the longer fence is safe under both. The decompiler accepts any matched fence of three or more backticks and warns when a log line can still close it early, which fails Step 6c condition (i) (notebook-assembler "Marimo: Check the Log Before Emitting Cell 3").
  • Trailing spaces and tabs on code or log lines do not survive the Write and Edit tools, and Step 6c cannot see the loss. Grep each final script for [ \t]$ and report every match as an INFO under Issues Found (notebook-assembler Step 6c, "What the two whitespace flags absorb").
  • No --sandbox. It runs the notebook outside the image's environment, and on marimo edit and marimo run it also rewrites the notebook file, adding a PEP 723 block (see ./references/apps-deployment.md > Docker with Sandbox).
  • marimo export html writes __marimo__/session/<notebook>.py.json beside the notebook every time (no opt-out); DAAF's .gitignore excludes __marimo__/.

Anti-Patterns (What BAD Output Looks Like)

# ❌ WRONG — This is new analysis code
tier_summary = risk_data.group_by("tier").agg(pl.len())

# ❌ WRONG — This is a dashboard widget
sector_dropdown = mo.ui.dropdown(options=["Public", "Private"])

# ❌ WRONG — This is a transformation when loading
df = pl.read_parquet("data.parquet").with_columns(pl.col("x") * 2)

# ❌ WRONG — This is filtering
filtered = df.filter(pl.col("state") == "VA")
# ✅ CORRECT — Contiguous canonical archive bundle
@app.cell
def _(mo):
    mo.md("""
    ### 5.1: Fetch Data

    **Script:** `scripts/stage5_fetch/01_fetch.py`
    **Status:** CP1 PASSED
    """)
    return

@app.cell
def _():
    # SOURCE: scripts/stage5_fetch/01_fetch.py
    # =========================================================================
    # ARCHIVED SCRIPT CODE (commented out to prevent execution conflicts)
    # Full executable script preserved at: scripts/stage5_fetch/01_fetch.py
    # =========================================================================
    #
    # import polars as pl
    #
    # output_path = "data/raw/file.parquet"
    # print(f"Saved existing output to {output_path}")
    pass  # Cell must have executable statement

@app.cell
def _(mo):
    mo.accordion({"Execution Log (01_fetch.py)": mo.md(r'''```
Executed: 2026-01-24 14:32:05
Exit code: 0
Checks: 5 passed, 0 warnings, 0 failed, 0 not verified, 0 skipped
CP1 VALIDATION: PASSED
```''')})
    return

# ✅ CORRECT — Either optional display form follows the complete bundle
# (PROJECT_DIR-based Path, one unique name per preview)
@app.cell
def _(pl, mo, PROJECT_DIR):
    df_5_1 = pl.read_parquet(PROJECT_DIR / "data/raw/file.parquet")
    mo.ui.table(df_5_1.head(100), selection=None, show_column_summaries=False)

@app.cell
def _(mo, PROJECT_DIR):
    mo.image(PROJECT_DIR / "output/figures/existing-figure.png")

See:

  • .claude/agents/notebook-assembler.md for the complete behavioral protocol
  • agent_reference/WORKFLOW_PHASE4_ANALYSIS.md Stage 9 for template

Quick Decision Trees

"I need to get started"

Getting started?
├─ Install marimo → ./references/quickstart.md
├─ Create first notebook → ./references/quickstart.md
├─ Understand how cells work → ./references/reactivity.md
└─ Run built-in tutorials → marimo tutorial intro

"I need interactivity"

Need interactive elements?
├─ Sliders, dropdowns, text inputs → ./references/ui-elements.md
├─ Tables with selection → ./references/ui-elements.md
├─ Forms with submit buttons → ./references/ui-elements.md
├─ Dynamic arrays of elements → ./references/ui-elements.md
├─ Charts with selection → ./references/sql-data.md (plotting section)
└─ Custom widgets (anywidget) → ./references/ui-elements.md

"I need to work with data"

Need data operations?
├─ Query dataframes with SQL → ./references/sql-data.md
├─ Connect to databases (Postgres, SQLite) → ./references/sql-data.md
├─ Display/filter dataframes → ./references/sql-data.md
├─ Create plots (Altair, Matplotlib, Plotly) → ./references/sql-data.md
└─ Interactive data exploration → ./references/sql-data.md

"I need to format output"

Need output formatting?
├─ Write markdown with Python values → ./references/outputs-layouts.md
├─ Arrange elements (hstack, vstack) → ./references/outputs-layouts.md
├─ Tabs, accordions, sidebars → ./references/outputs-layouts.md
├─ Progress bars, spinners → ./references/outputs-layouts.md
└─ Conditional output display → ./references/outputs-layouts.md

"I need to share/deploy"

Need to share or deploy?
├─ Run as read-only web app → ./references/apps-deployment.md
├─ Execute as Python script → ./references/apps-deployment.md
├─ Export to HTML/WASM → ./references/apps-deployment.md
├─ Deploy with Docker → ./references/apps-deployment.md
├─ Grid/slides layout → ./references/apps-deployment.md
└─ Convert from Jupyter → ./references/quickstart.md

"Something isn't working"

Having issues?
├─ "Multiple definitions" error → ./references/gotchas.md
├─ Cells not re-running as expected → ./references/reactivity.md
├─ UI element not updating → ./references/gotchas.md
├─ Cycle dependency error → ./references/gotchas.md
├─ Expensive cells running too often → ./references/gotchas.md
├─ Mutations not triggering updates → ./references/reactivity.md
├─ Lint a notebook without running it (marimo check) → ./references/quickstart.md
└─ Check that every cell executes (export html) → ./references/apps-deployment.md

Quick Reference

Essential CLI Commands

CommandPurpose
marimo editLaunch notebook server
marimo edit notebook.pyCreate/edit specific notebook
marimo run notebook.pyRun as read-only app
python notebook.pyExecute as script
marimo tutorial introRun interactive tutorial
marimo convert nb.ipynb -o nb.pyConvert other formats INTO marimo (inbound only)
marimo check notebook.py --format jsonLint without running (--select MB for breaking rules only)
marimo export html notebook.py -o out.htmlRun every cell and export to HTML; exits 1 if any cell fails

Docker: When running in a container, add --host 0.0.0.0 --port 2718 --headless to run and edit commands.

Core marimo Library

FunctionPurpose
mo.md("text")Render markdown
mo.ui.slider(start, stop)Create slider
mo.ui.dropdown(options)Create dropdown
mo.ui.table(data)Interactive table
mo.hstack([...])Horizontal layout
mo.vstack([...])Vertical layout
mo.sql(f"SELECT ...")SQL query
mo.stop(condition)Stop this cell and the cells that read its names

Topic Index

TopicReference File
Installation & Setup./references/quickstart.md
CLI Commands./references/quickstart.md
Linting (marimo check, rule codes)./references/quickstart.md
Reactive Execution./references/reactivity.md
Cell Dataflow./references/reactivity.md
Validation Patterns./references/validation-patterns.md
Transform-Validate Pairs./references/validation-patterns.md
Data Quality Checkpoints./references/validation-patterns.md
Sliders & Inputs./references/ui-elements.md
Tables & Forms./references/ui-elements.md
SQL Cells./references/sql-data.md
DataFrames./references/sql-data.md
Plotting./references/sql-data.md
Markdown./references/outputs-layouts.md
Layouts./references/outputs-layouts.md
Apps & Scripts./references/apps-deployment.md
marimo run Is Not an Execution Test./references/apps-deployment.md
Export & Deploy./references/apps-deployment.md
Session snapshots (__marimo__/)./references/apps-deployment.md
Access tokens (DAAF notebook browser; keep tokenized links out of logs and notes)./references/apps-deployment.md
Common Errors./references/gotchas.md
Best Practices./references/gotchas.md

Citation

When this library is used as a primary analytical tool, include in the report's Software & Tools references:

Agrawal, A., & Scolnick, M. (2023). marimo - an open-source reactive notebook for Python [Computer software]. https://doi.org/10.5281/zenodo.12735329

Cite when: The analysis notebook is delivered as a marimo notebook (typically always true in DAAF pipelines). Do not cite when: marimo is not used for the analysis delivery format.

writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 319kUpdated Claude CodeCodexVerifiedInstall
xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.Stars: 196kUpdated Claude CodeCodexVerifiedInstall
shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy.Stars: 118kUpdated Claude CodeCodexVerifiedInstall
childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release.Stars: 117kUpdated Claude CodeCodexVerifiedInstall