Skip to main content

Task Planning

The problem​

An AI agent needs a structured task plan — steps with dependencies, inputs, outputs, and conditions. Not a wall of text, but something a machine can execute and a human can audit.

The solution​

title: Customer Onboarding Pipeline
summary: Automated pipeline for new enterprise customer onboarding
meta: | type: agentic | domain: operations

section: Data Collection

step: Gather customer information | id: gather | tool: crm-lookup | input: customer_id | output: customer_data | status: pending
step: Validate business license | id: validate | depends: gather | tool: license-api | input: customer_data.license_number | output: validation_result | status: pending
step: Credit check | id: credit | depends: gather | tool: credit-api | input: customer_data.tax_id | output: credit_score | confidence: 0.95 | status: pending

section: Evaluation

decision: Credit approved? | if: credit_score > 650 | then: setup | else: manual-review
decision: Account tier | options: standard, premium, enterprise | input: customer_data.revenue | output: tier

section: Account Setup

step: Create account | id: setup | depends: validate, credit | tool: account-api | input: customer_data, tier | output: account_id | status: pending
step: Configure permissions | id: permissions | depends: setup | tool: iam-service | input: account_id, tier | status: pending
step: Send welcome email | id: welcome | depends: permissions | tool: email-service | input: customer_data.email, account_id | status: pending

section: Monitoring

trigger: Onboarding complete | event: welcome.sent | signal: onboarding-complete
audit: Pipeline initiated | by: onboarding-agent | at: 2026-03-15T10:00:00Z | level: info

done: Onboarding pipeline ready | status: template | time: 2026-03-15T10:00:00Z

Step by step​

  1. Steps — Each step: is a unit of work with an id:, optional depends: for ordering, tool: for execution, and input:/output: for data flow.
  2. Gates — gate: blocks pause execution until a named approver grants permission. Use decision: for automated condition-based branching.
  3. Decisions — decision: blocks select from options based on data.
  4. Triggers — trigger: fires on events. signal: sends events for other pipelines.
  5. Audit — audit: blocks log what happened and when.
  6. Done — done: marks the pipeline as complete or ready.

Querying the plan​

# All steps and their status
intenttext plan.it --query "type=step"

# Pending steps only
intenttext plan.it --query "type=step status=pending"

# Steps that depend on a specific step
intenttext plan.it --query "type=step depends:contains=gather"

Next steps​