cubenlp/skills/practice-make-perfact

Practice Make Perfact

Use this skill after a ChatTool feature, fix, refactor, CLI change, skill update, or documentation-backed behavior change is basically working.

This is a post-task normalization workflow, not a pre-task exploration guide.

Workflow

1. Review the actual work product

   • Read the current diff, changed files, and any generated outputs.
   • Separate durable repository artifacts from scratch notes or temporary experiments.
   • Keep temporary exploration outside the repository unless it must become maintained documentation.

2. Update project-local records first

   • If the task lives under projects/<date>-<name>/, update that project's PRD.md, progress.md, and memory.md before promoting durable artifacts.
   • Do not promote raw one-off research into workspace root references.
   • Move only normalized, durable outputs into ChatTool src/, docs/, skills/, or tests/.

3. Check existing CLI surfaces before adding scripts

   • Look at references/cli-reference.md first.
   • Prefer existing chattool pypi, chatpypi, chattool skill, chatenv, chattool gh, and chattool dns commands over ad hoc scripts.
   • Add a script only when the action is repeatable, fragile, and not already covered by a CLI.

4. Normalize ChatArch boundaries

   • ChatTool commands should use chattool.interaction adapters when they need ChatTool policy or test patch points.
   • Standalone ChatArch packages should use external chatstyle and chatenv directly.
   • New env/profile schemas should be provider modules registered through chatenv.configs.

5. Normalize skills

   • Skill changes must keep SKILL.md and SKILL.zh.md aligned.
   • If a skill has agents/openai.yaml, update the metadata when behavior or trigger scope changes.
   • New skills should normally include English and Chinese instructions plus agents metadata.
   • Run a stale-command grep after skill edits.

6. Run the post-task review

   • Review the current diff, not the whole repo by default.
   • Check lazy imports, interactive -i/-I behavior, ChatStyle/ChatEnv boundaries, docs, tests, README, and changelog.
   • Use $chattool-dev-review unless the user explicitly says not to.

7. Carry the work to PR/MR stage

   • Commit and push the branch.
   • Use chattool gh for GitHub PR operations.
   • Prefer chattool gh pr create --body-file ... and update the PR body when scope changes.
   • Use chattool gh pr checks --wait when CI status is part of the handoff.

8. Hand off real release work

   • Tags, package publishing, PyPI verification, and release.log belong to $chattool-release after merge.
   • Do not tag from an unmerged PR head.

Required Outputs

• The repository diff is normalized into maintained locations.
• Project-local PRD.md / progress.md / memory.md are updated when present.
• Skills include matching English and Chinese files.
• Docs/tests/changelog updates accompany behavior changes.
• The task reaches PR/MR stage unless the user explicitly asks to stop earlier.