Codex App Explain

Use another workflow when the user is primarily asking to:

• update a persistent preference or memory
• find a file or source without requesting explanation
• write or complete a full assignment
• review code or documents

If the request is mixed, locate the source first and then run this explanation workflow on the exact object to be explained.

Read Order

Read only the minimum context needed for the current blocker.

1. The host environment's active answer-format or notation rules, if available
2. The user's cited source material or local project material
3. This skill's bundled references when they help disambiguate the workflow

Use references/explanation-modes.md when you need the public mode split for:

• single-concept explanation
• compact prerequisite-chain rebuild
• sentence-by-sentence paper reading
• full layered teaching

For topic material:

• Locate the cited source using available local or project retrieval before explaining.
• Load the smallest exact fragment that supports a rigorous explanation.
• Prefer one page, one equation, one excerpt, or one bounded problem part when that is enough.

Core Workflow

1. Bind the Object and Scope

Before explaining, explicitly bind:

• what is being explained
• which source fragment it comes from
• which scope applies to it

Name the object first. Then name the source. Then state the scope boundary.

If a missing prerequisite blocks the explanation, insert the minimum prerequisite object first and say why it is being introduced.

2. Load Minimal Evidence

Load only the exact material needed for the current blocker.

Prefer:

• the page, excerpt, or equation under discussion
• the smallest notation source that resolves the ambiguity
• one bounded source fragment rather than a whole document

3. Choose Explanation Depth

Choose depth from the user's current request.

Default behavior:

• Give the full layered explanation when the topic is new or the user asks for a step-by-step rebuild.
• Answer only the exact blocker when the follow-up targets one local gap.
• Re-answer the same topic directly when the previous version failed on clarity, structure, or formatting.

Public mode split:

• short-concept mode
  • Use for one concept, one term, one symbol family, or one local blocker.
  • Keep the answer local: definition first, one minimal example when needed, then intuition or application only if it still helps.
• prerequisite-chain mode
  • Use when the user's real blocker is not one sentence or one formula, but a short hidden dependency chain such as state -> trajectory -> flow map -> observable.
  • Name the hidden chain, rebuild it in dependency order, then stitch back to the source.
• paper-sentence mode
  • Use only when the user explicitly asks for sentence-by-sentence or passage-by-passage reading of a paper, PDF, or note.
  • Preserve the source sentence visibly, then explain it in a low-barrier pass and a deeper interpretation pass.
  • When the quoted source contains OCR noise or poor ASCII-style symbol rendering, apply minimal Unicode-friendly cleanup to improve readability.
  • Keep that cleanup display-only: preserve the original English wording and meaning, and do not paraphrase, translate, or insert explanation inside the quote.
  • Add a local concept note only when that sentence contains a true blocker.
  • Optional paragraph-level and section-level synthesis may follow after the relevant local blocks.
• full-layered mode
  • Use for full derivations, from-basic-to-rigorous rebuilds, or when the topic is new enough that a compact answer would hide the real bridge steps.

4. Deliver the Explanation

Deliver the explanation in a deterministic layered order when full depth is