Discovery, research, content, and health workflows

Product Discovery

Type: discovery
Primary output: scored and validated product candidates
Human gates: iteration and candidate approval

Discovery turns business context into candidate opportunities. Candidates can be inspected, scored, selected, approved, and elaborated into more actionable work.

Configuration can include a discovery scope, candidate limit, team/project target, notification channel, auto-approval policy, and downstream implementation agent. The run detail shows discovery, evaluation, validation, iteration, approval, task creation, and notification progress.

Use the main Discovery pipeline view for candidate approval and elaboration actions.

Elaboration

Type: elaboration
Primary output: implementation-ready product scope or ticket
Trigger: approved Discovery candidate

Elaboration converts a selected candidate into personas, use cases, edge cases, error states, acceptance criteria, and success metrics. After ELABORATING, it waits at AWAITING_REVIEW; approval can create a project-management ticket and optionally trigger the Developer workflow.

Elaboration is a Discovery follow-on rather than a general recurring-schedule choice. Preserve the originating candidate ID when diagnosing a broken Discovery-to-Developer chain.

Research

Type: research
Primary output: report, analysis, brief, or recommendation
Human gate: review the evidence before action

Research accepts a topic plus an optional scope (competitive_analysis, market_research, user_research, or technical_research) and deliverable type. It gathers authorized venture context, starts an isolated execution environment, researches, then synthesizes the result.

Always verify consequential factual claims and source recency before acting on a report.

Content

Type: content
Primary output: reviewable channel-specific draft
Human gate: editorial approval before publication

Content accepts a brief, content type, target audience, and optional tone, length, SEO keywords, research phase, agent, repository, and resource context. Supported content types include landing page, blog post, email sequence, social media, ad copy, product description, and other.

The workflow sets up the run, gathers context, starts execution, and drafts the content. When enabled, research precedes drafting. A repository can be mounted for technically accurate product content. The output enters review; publication is a separate authorized operation.

See Content review and public blog for the delivery state machine.

Health Check

Type: health_check
Primary output: diagnostic findings and report
Health Check accepts full, metrics_only, incidents_only, or custom scope, plus an optional lookback window and named alert thresholds. It gathers available venture and operational context, runs diagnosis, and emits a report.

This is a diagnostic workflow, not a direct probe of every external service and not a replacement for provider-native monitoring. A clean result means no issue was found in the supplied signals.

Department Head

Type: department_head
Primary output: reviewed escalations and dispatched specialist work
Trigger: schedule or autonomous department operation
The Department Head workflow gathers department context, prepares Brain knowledge, starts its execution environment, generates or loads its working personality, investigates unknowns, reviews prior task outcomes and escalations, then dispatches work. Child runs retain a parent workflow reference so operators can follow delegation.

Unlike a specialist workflow, Department Head is an orchestrator. Its output may be more workflow runs rather than a single artifact. Resource contexts must be pre-authorized before dispatch; the agent cannot gain credential access simply by naming a resource.

Generic Task

Type: generic_task
Primary output: freeform result from the selected agent
Trigger: one-off manual or agent dispatch
Generic Task handles work that does not fit a typed primitive. It is intentionally absent from recurring workflow choices. Prefer a typed workflow when one exists: typed configuration, phase semantics, retry behavior, outputs, and analytics are more predictable.

Reliability and recovery

Every workflow exposes durable state and error reporting. The exact recovery actions depend on the workflow and current phase; use the actions shown on the run detail, and contact support when an interrupted run does not settle.