Brain orchestrator skill. Runs the knowledge-gathering loop: pre-analysis, standup, role chain execution, retrospective, and knowledge area updates. Use before and after every task, or when the user asks to "run brain".
Brain wraps the entire role chain in a knowledge loop. It ensures every role has the context it needs and that knowledge is never lost between tasks.
Brain Pre-analysis → Standup → Role Chain → Retrospective → Knowledge Update → Loop
Gather all existing knowledge and understand the current state of the codebase relevant to the task.
Read ALL files in .cursor/context/:
architecture.md — system structurepatterns.md — coding conventionscomponent-inventory.md — DS catalogtech-debt.md — known issuesdecisions.md — past ADRsmetrics.md — quality numbersproduct.md — product contextExplore the codebase areas relevant to the task:
Identify knowledge gaps:
A mental model of the current state that feeds the standup.
Every role reports its perspective. Brain synthesizes into a Task Brief.
Run through each role's domain and generate a report:
patterns.md apply?patterns.md that cover this use case?component-inventory.md are relevant?dry-audit.sh currently report?patterns.md apply?tech-debt.md)?metrics.md?After all roles report, Brain produces:
TASK BRIEF — [Task Name]
Date: [date]
CONTEXT:
[Summary of what Brain knows about the affected area]
KEY DECISIONS:
[List of decisions that need to be made during execution]
REUSE OPPORTUNITIES:
[Components, hooks, patterns that should be reused]
RISK AREAS:
[Things that could go wrong or need extra attention]
KNOWLEDGE GAPS:
[What Brain couldn't determine and needs to discover during execution]
The normal role chain runs (Phases 1-9 from web-workflow/SKILL.md), but each role:
Capture everything learned during execution and update knowledge areas.
Review changes made: what files were created, modified, deleted.
Update knowledge areas:
architecture.md — update if:
patterns.md — update if:
.cursor/rules/component-inventory.md — update if:
tech-debt.md — update if:
decisions.md — update if:
metrics.md — update if:
product.md — update if:
Evaluate role effectiveness:
Determine if another loop is needed:
Updated .cursor/context/ files and a determination of whether to loop again.
When the user says "run brain", "standup", or "analyze":
This is the "deep learning" mode where Brain builds complete knowledge from scratch.
Each file in .cursor/context/ follows this structure:
# [Area Name]
> Last updated: [date]
> Updated by: Brain retrospective after [task name]
## [Section]
[Content organized by topic, with specific file paths and details]
Keep entries specific, actionable, and linked to real file paths. No vague summaries.