Skip to content

Build Claude Code-style intent understanding UX for Codexplain #13

@minsing-jin

Description

@minsing-jin

Goal

Build a Codexplain intent-understanding layer that makes Codex explanations feel closer to Claude Code UX: the system should infer what the user really wants, choose the right explanation form, and avoid dumping file-level or micro-step prose when a conceptual answer is needed.

Why

Current Codexplain improves rendering, width safety, row dividers, color, progress UI, and native TUI integration. The next UX gap is intent understanding: users ask in natural language, and the tool should reliably decide whether the answer needs architecture, diagnosis, comparison, workflow progress, risk review, next actions, or a short direct answer.

Requirements

  • Detect user intent from one message and recent context: architecture, problem diagnosis, progress report, comparison, decision, implementation plan, research-style explanation, cause-effect report, and direct Q&A.
  • Prefer capability/function-level explanations over file-by-file summaries unless the user explicitly asks for files.
  • Use answer-first structure, then flow downward into evidence, reasoning, fixes, and next actions.
  • Keep Korean responses Korean and default open-source docs/examples English.
  • Route to existing Codexplain renderers: TLDR, wide-divider table, flow boxes, progress bar, risk panel, problem-diagnosis, cause-effect, Notion-style blocks, and checklist.
  • Add a small explainable decision trace for debugging, without polluting normal answers.
  • Preserve strict-output safety for JSON, code, diffs, logs, tests, patches, and commit messages.
  • Provide project-local on/off controls and no global Codex mutation by default.

Acceptance Criteria

  • codexplain shape chooses the intended renderer for at least 20 representative prompts across Korean and English.
  • Architecture prompts produce concept/function-first diagrams, not file-first prose.
  • Problem prompts produce answer-first diagnosis without forcing rigid labels when the language/context calls for a more natural flow.
  • Long tables and diagrams remain width-safe under quality-check --width 88.
  • A debug command shows why an intent/renderer was selected.
  • cargo test, codexplain quality-check --width 88, and codexplain compat-check pass.

Notes

This should be implemented as a Codexplain UX layer, not as global Codex configuration. Native Codex TUI hooks should remain reversible through the project-local adapter.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions