procedure

Agent lint and dry-run

Before an agent proposes code, it should lint (and dry-run or execute when available) the same way the Code tab does — so you are not approving broken scripts.

This page lists the validation tools per agent and what you should expect to see in chat when something fails.

Per tier

Tier	Lint	Dry-run / execute	Propose
Indicator	`lint_indicator` (client)	`dry_run_indicator` (server)	`propose_indicator_edit`
Definition	`lint_flow` over live draft	`dry_run_flow` removed from agents	`propose_flow_edit` patches
Study	study lint paths in builder	`execute_study_run`	`propose_study_edit`
Pins	rewrite validates pin contract	study re-run after edit	`propose_chart_pin_edit`

What humans see

Lint errors appear in chat tool results and in the Code tab diagnostics.
Propose tools require your approval before code lands.
Budget / dependency errors often show only on dry-run or full run — lint is necessary but not sufficient.

Checklist before propose

Declarations match emits / publishes.
Labels present on every ctx.emit.
No module-level bar state.
Dependency contracts resolved.
For pins: accepted mockup + additive pin_cell only.

← Previous Next →