quarto-qmd-cli-rendering

Goal

When asked to “render”, “build”, “preview”, or “compile” a Quarto .qmd, respond with quarto CLI commands that are copy-pastable and include brief, practical notes.

Include a short note that the terminal will show per-cell progress lines like:

Cell 1/3 '{cell-label}'.....Done
Cell 2/3 '{cell-label}'.....Done
Cell 3/3 '{cell-label}'.....Done

Output created: {file-name}

Some cells finish quickly while others can take a long time; the agent should be patient and wait for completion. For example, the following terminal indicates that cell 2 is still running:

Cell 1/3 '{cell-label}'.....Done
Cell 2/3 '{cell-label}'.....

Assumptions

• Quarto CLI is installed and available as quarto.
• The user can run commands from a terminal in the relevant directory, or will provide an explicit path.
• The target output format is either defined in YAML, or the user will specify it.

Core rules

• Single-file render: use quarto render <file.qmd>.
• Project/directory render: use quarto render or quarto render <dir> (named project directory).
• Prefer explicit success signaling in scripts, but match the user's shell:
  • PowerShell: append ; if ($LASTEXITCODE -eq 0) { echo "Render finished" }
  • bash/sh: append && echo "Render finished"
• If the user says “render sequentially” or “low RAM”, render one file at a time (do not suggest parallel execution).
• If a render fails, use the failing cell label and traceback to localize the issue to the specific chunk before making edits.
• When the terminal appears stuck on one notebook cell, do not assume a hang immediately; first confirm whether the Python kernel or model-artifact directory is still active.

Render a single .qmd

Minimal pattern:

quarto render <file.qmd>

Optional explicit success signal (choose ONE depending on shell):

quarto render <file.qmd> ; if ($LASTEXITCODE -eq 0) { echo "Render finished" }

quarto render <file.qmd> && echo "Render finished"

Sequential rendering (low RAM)

Use sequential rendering when the user requests:

• “One at a time.”
• “No parallelism.”
• “Low RAM” or “don’t run everything at once.”
• “Render each qmd and stop on error.”

Instructions:

1. Render exactly one file:
   • PowerShell: quarto render <file.qmd> ; if ($LASTEXITCODE -eq 0) { echo "Render finished" }
   • bash/sh: quarto render <file.qmd> && echo "Render finished"
2. Wait until the command exits (i.e., when the terminal shows "Render finished") before proceeding to the next file.
3. If it fails, fix the current .qmd, then re-run the same command until it succeeds.
4. Continue to the next .qmd only after the current one succeeds.
5. The command succeeds when there are no error messages in the terminal and you see Render finished.

Debugging a failed render

When Quarto reports a failing cell:

1. Read the traceback and capture the chunk label (for example, Cell 47/124: 'calibration-s2').
2. Read only the nearby chunk in the .qmd plus one neighboring chunk that implements the same pattern.
3. Form one local hypothesis about the failure from the exact traceback, such as an incomplete function call or a shadowed imported helper.
4. Fix the smallest local defect first, then rerun the full single-file render.
5. If the file contains repeated strategy/model sections, search for sibling chunks with the same structure before rerendering so the next copy does not fail later.

Prefer local chunk comparisons over broad repo exploration; most render failures are plain code or markdown defects inside the current .qmd.

Skill scope

Use this skill when:

• The user mentions .qmd, Quarto, “render/build/compile/preview”, CI, Makefiles, or scripting.

Do not use this skill when:

• The user explicitly wants GUI-only steps (RStudio/VS Code) with no CLI commands.

Troubleshooting common rendering errors

Error: "Unable to locate an installed version of Python 3"

Quarto looks for python3 on the PATH, but conda environments only provide python (not python3). Fix by setting the QUARTO_PYTHON environment variable to the full path of the Python executable before running quarto render.

First, find the Python path:

conda activate <env_name>; python -c "import sys; print(sys.executable)"

Then use it when rendering:

conda activate <env_name>; $env:QUARTO_PYTHON = "<full-python-path>"; quarto render <file.qmd>; if ($LASTEXITCODE -eq 0) { echo "Render finished" }

Example:

conda activate env_AutoGluon_202502; $env:QUARTO_PYTHON = "C:\Users\chiuw\miniforge3\envs\env_AutoGluon_202502\python.exe"; quarto render Labs/Lab_01.qmd; if ($LASTEXITCODE -eq 0) { echo "Render finished" }

Error: "Unknown project cache version" / "Assertion failed" (Sass/revealjs)

This occurs when Quarto's internal cache is stale or was created by a different Quarto version. The error often manifests during Sass compilation for revealjs themes.

Fix: Clear all Quarto cache directories recursively before rendering:

conda activate <env_name>; Get-ChildItem -Recurse -Directory -Filter "_quarto*" | Remove-Item -Recurse -Force -ErrorAction SilentlyContinue; Get-ChildItem -Recurse -Directory -Filter ".quarto*" | Remove-Item -Recurse -Force -ErrorAction SilentlyContinue; Get-ChildItem -Recurse -Directory -Filter "_site*" | Remove-Item -Recurse -Force -ErrorAction SilentlyContinue; echo "All caches cleared"

Then retry the render with --no-cache:

conda activate <env_name>; $env:QUARTO_PYTHON = "<full-python-path>"; quarto render <file.qmd> --no-cache; if ($LASTEXITCODE -eq 0) { echo "Render finished" }

Error: "PermissionDenied: Access is denied. (os error 5)" during cleanup

Harmless cleanup error. Use -ErrorAction SilentlyContinue on Remove-Item commands to suppress it. Does not affect render output.

Symptom: render appears stuck on one cell for a long time

This is common for notebook-backed .qmd files that train models or generate large reports. A terminal line such as:

Cell 29/124: 'lab04-fit-autogluon-s1'...............

does not imply a hang by itself.

Before treating it as stalled:

1. Check whether the shell prompt has returned. If not, the render is still running.
2. Check whether the Python kernel process is still alive and consuming CPU or memory.
3. If the cell writes model artifacts, check whether files in the output directory are still getting fresh timestamps.

If those checks show activity, keep waiting. Long AutoGluon or SHAP cells can legitimately run for many minutes.

Symptom: a late-cell failure after a long successful run

Do not restart broad debugging. Use the exact failing cell and traceback to inspect the local chunk first.

Common causes in .qmd code cells:

• an incomplete function call or missing closing delimiter in one repeated chunk
• a variable name that shadows an imported callable such as display
• copying one strategy block and forgetting to finish or rename one line in the sibling block

General pre-render checklist

Before rendering any .qmd in this project, always:

1. Activate the conda environment: conda activate env_AutoGluon_202502
2. Set $env:QUARTO_PYTHON to the conda env's Python executable
3. Clear stale Quarto caches if previous renders failed
4. Use --no-cache if cache issues persist
5. Append the success signal: ; if ($LASTEXITCODE -eq 0) { echo "Render finished" }

Render time expectations

• revealjs slides (Lectures): Fast — typically under 30 seconds for slides with no executable code cells
• HTML labs with AutoGluon training: Slow — can take 5–30+ minutes depending on time_limit settings. Cells that train models (e.g., TabularPredictor.fit()) are the bottleneck
• HTML labs without model training: Moderate — typically 1–5 minutes for EDA and plotting cells

Always use mode='async' when running render commands and poll with get_terminal_output until the success signal appears.