| name | scout |
| description | Ground an unfamiliar repo area, library, API, failure mode, or design question. Writes evidence to .context/scout.md and returns only constraints, risks, unknowns, and the next natural action. |
/scout
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
Map facts that would change a later decision.
Inputs
A concrete topic from the user, current repo context, an issue/PR, or the active uncertainty. If the topic is vague, inspect the repo first, then ask one clarifying question only if needed.
Reads
Repo code/tests/docs, git history, installed package source/types, official docs or context7 for library/API behavior, gh_grep/reference repos for usage, exa for current external pages.
Writes
.context/scout.md.context/session-state.md
Process
- State the scout question in one sentence.
- Gather only evidence that could change boundaries, failure model, state model, API shape, security, migration, or next questions.
- Separate facts from assumptions.
- Put source notes, snippets, raw findings, and rejected leads in
.context/scout.md.
- Update session state with constraints, unknowns, evidence path, and next action.
Operator Output
Return: decision, up to three evidence bullets, risk/unknowns, and next action. Suggest /interview when intent needs pressure; suggest /architect when the problem and direction are already clear.
Stop Conditions
Stop when new evidence is unlikely to change the next decision, or when one user decision blocks further grounding.
