concept

Overview

Authoring agents write the same Code-tab scripts a human would — Conductor routes your request to the right specialist instead of guessing across indicators, definitions, studies, and pins.

Use this map when you are unsure which chat agent owns a change, especially Put on chart (Forward insight agent, not the study authoring agent).

Who does what

Agent	Authors	Does not author
Indicator authoring	Indicator Code (`meta`, `output.*`, `onBar`)	Definitions, studies, pins
Definition authoring	Definition Code (events, emit, modules)	Studies, pins
Study authoring	Study Code (flows, collect, publish)	`pin_cell` / `popover_layout` for Put on chart
Forward insight agent (+ rewrite)	Mockups → `propose_chart_pin_edit`	Research study logic (keeps it additive)
Conductor	Delegates + product steps	Raw cross-tier code dumps

Routing map

You / Conductor
  ├─ chart math → indicator authoring
  ├─ mark setups → Definition authoring
  ├─ measure outcomes → study authoring
  └─ put stats on chart → Forward insight agent (mockup → rewrite)

Shared rules for agents

Lint before propose.
Resolve dependency contracts before inventing deps.
Prefer patches / full-file proposes the product expects for that tier.
Pause for human approval on edits that change saved code.
Never treat pin authoring as part of study authoring.

Pages in this section

Page	Focus
Indicator authoring agent	Tools + prompting
Definition authoring agent	Multi-file patches
Study authoring agent	Runs + no pins
Forward insight agent	Mockups → pin_cell
Agent lint and dry-run	Quality gates

← Previous Next →