Help the learner repair a plotting function using code, tests, and a chart made from invented summary counts.
Read README.md, REPO_MAP.md, and the relevant page in docs/paths/python/ or docs/paths/r/. The plotting contract is docs/reference/io_contract.md. Python uses plotting/plot_summary.py; R uses plotting/plot_summary.R.
The starting chart deliberately overlaps two groups and clips long labels. Preserve that unfinished exercise during maintenance unless the user asks to solve it. For a learner's repair, change the plotting function to make the grouped chart follow docs/reference/figure-specifications.md while preserving categories, counts, group assignments, and order.
Make the smallest clear change that meets the figure requirements and that a Python or R researcher can explain and maintain. Work in the existing plotting function. Prefer direct code and the current dependencies; justify any new dependency, configuration, class, helper, or extra processing by a concrete need in this task. Do not turn a figure repair into an unrequested production pipeline or reusable framework. If the task appears to require broader work, explain the scope increase and a simpler alternative and wait for the learner's decision before expanding the scope.
Simplicity includes readable names and useful comments, validation, and accessibility. Do not remove these to reduce line count or compress code into clever expressions. Avoid repeated work and unnecessary computation; do not optimize hypothetical future workloads. Passing tests does not establish maintainability or efficiency. Support performance claims with measurements or a specific explanation of the operations involved, and label estimates as estimates.
Explain source files, function calls, and tests. Do not open individual-level data, inspect participant records, or diagnose why a data point failed. This exercise has no participant records or data-file inputs. Its fixed aggregate counts are invented and embedded in source code.
Use the supplied chart to discuss labels, spacing, axes, and readability. Run the synthetic checks when requested and use their pass/fail evidence. Tests do not establish visual quality; inspect the rendered image separately. Do not change expected values to make a test pass.
When reusing this workflow, work in a separate workspace with code permitted for the selected service and invented examples. Keep study data, exports, logs, credentials, and sensitive code literals outside the agent's access. A local terminal, read-only mode, or .gitignore is not a data-sharing permission. See docs/reference/lab-data-policy.md.
Follow the chosen client and the learner's current stage. The default walkthrough uses VS Code with Copilot's Local Ask → Plan → Agent roles, Explorer, and Source Control; complete terminal instructions are in docs/appendix/command-line.md. Explanation, planning, and verification-only requests do not authorize repairs.
Use docs/reference/agent-control.md for permission and independence guidance. Explain the purpose, affected files or network access, and duration of a permission request without encouraging automatic approval. Honor existing authorization; do not broaden access, disable safeguards, or change global settings to bypass a failure. Instructions and skills do not enforce a sandbox. Leave the learner able to explain and maintain the code: provide checkable reasoning, actual evidence, and the requested opportunity to review before proceeding.
From the repository root, run the chosen language's checks:
python -m unittest discover -s plotting/testsRscript plotting/tests/run_tests.RAlso run the selected language's separate figure checker as described in docs/appendix/command-line.md. Its rejection of the unfinished starter is intentional; a completed repair must pass both behavior and figure checks against the latest source. Verification-only requests must report failures without making further repairs.
Render a new image, compare it with the saved baseline, and inspect the code diff. Follow the journal's accessibility requirements: review contrast, legibility, and group identification without color, and write accurate alternative text in runs/with-fix/summary.alt.txt. Attach that text when placing the image in a document or slide. Consider accessibility whenever creating an artifact for others to use. Keep runs/ and tmp/ ignored. Use .agents/skills/plot-review/SKILL.md for a requested independent review. The optional lab-r-refactor skill applies the local R template without changing behavior.
Preserve unrelated work and agree on file ownership with other agents. Finish with what changed, actual commands and outcomes, visual observations, and anything untested. Do not publish or modify external services without authorization.