auto_task.create_specs_from_todos

Goal

• Given a list of TODOs provided by the user in the form of a cfile create a plan to fix them

Workflow

Understand Each TODO

• Each TODO is in the form

  <FILE>:<LINE_NUM>:<TODO description>
  

  • E.g.,

    ./helpers/hmarkdown_coloring.py:235:def colorize_bullet_points_in_slide(...
    

• Read each TODO and understand the problem
• If it's a bug:
  • Make sure to understand the root cause
  • Run experiments to verify that you understood the issue and its root cause
• Devise a plan to fix the problem
• Devise a plan to verify that the problem is fixed

Write the Fix Plan

• Create a plan tasks.md for fixing the tests using the template below .claude/templates/auto_task.template.md
• Fill in the template's * Repo: checklist per .claude/skills/auto_task.rules.md section "Multi-Repo Issues, Branches, and PRs" (how to tell which repo a <FILE> belongs to, and how to handle TODOs spanning more than one repo)
• Write the comments as bullet points according to .claude/skills/markdown.rules.md and .claude/skills/text.rules.md with minimal text
• Do not make any change to the code, but only propose the fixes

Follow the Conventions

• Follow .claude/skills/auto_task.rules.md section "Follow the Coding and Testing Rules"
  • Refer to skills in .claude/skills/coding.*/SKILL.md and .claude/skills/testing.*/SKILL.md to follow the repo's best practices
• For bugs, add unit tests to check that the problem was present and then it's fixed (red/green approach)
• For renaming, make sure the entire code base is checked to make sure everything is updated
  • Follow .claude/skills/coding.rename/SKILL.md
• For change of function signature, make sure all the calling instances have been updated

Rank the Issues

• Reorder the issues in increasing complexity, starting from the issues with high confidence in the fix

Verification

• [ ] tasks.md contains one issue block per input TODO
• [ ] Every issue block has type, reason, proposed fix, confidence, complexity, and verification plan filled in
• [ ] Issues are ordered by increasing complexity, starting from the highest confidence fixes
• [ ] No code was changed, only the plan file was written