| name | teach |
| description | Teach a concept, solution, or pattern one layer at a time with grounded examples. Uses the attention contract and avoids dumping full derivations unless requested. |
/teach
Read the shared contracts before output: ../_shared/operator-output.md, ../_shared/attention-budget.md, ../_shared/evidence-record.md, ../_shared/state.md, ../_shared/cognitive-load.md, ../_shared/collaboration.md.
Purpose
Make one concept usable without overwhelming the learner.
Inputs
Named concept, recent architecture/fix/design, repo code, library/API question, or user request to learn.
Reads
Actual repo code for repo topics, official docs/context7/package source for library/API behavior, and .context/ artifacts when teaching a recent workflow result.
Writes
.context/teach.md when the explanation depends on evidence or should persist.context/session-state.md for substantial teaching sessions
Process
- Identify the one concept being taught.
- Ground facts before explaining repo/API behavior.
- Teach the first useful layer only: problem, simple model, concrete example, where it fails.
- Ask at most three check questions or offer the next layer.
- Put extended notes or source details in
.context/teach.md when needed.
Operator Output
Use compact teaching prose, not the workflow Decision template unless a decision is being taught. Include one concrete example and the next learning step.
Stop Conditions
Stop if the topic is ambiguous and cannot be inferred, or if deeper layers require user choice.
