Runbook 09.5 — URL → Test Plan Agent (Discovery Layer)

Quick links

Navigation

Use as a checklist.

Objective Done Inputs Mechanics Commands Expected Failure modes Operational value

Objective

Given a target URL, produce a deterministic, structured baseline test plan.
Extract visible navigation and page structure into a capability-oriented inventory.
Generate a first-pass locator map aligned to semantic, user-facing elements.
Establish the discovery layer that later agents can consume.

Rule

This sprint is not autonomous test generation. It is deterministic discovery from a real system entry point.

Definition of done

✔ Same URL produces the same structured plan on repeat runs.
✔ At least one capability is identified from real navigation or page structure.
✔ At least one smoke candidate is identified.
✔ A locator map is generated alongside the plan.

Inputs + outputs

Input URL	Target system entry point such as `http://sut.testlab:3000`
Execution mode	Deterministic page load, no uncontrolled crawl
Primary outputs	`test-results/agent-output/test-plan.json` and `test-results/agent-output/locator-map.json`

Example input

{
            "url": "http://sut.testlab:3000",
            "env": "qa"
          }

Mechanics

Navigate to the provided URL in a known browser state.
Capture links, forms, and major entry-point controls.
Map visible system structure into candidate business capabilities.
Assign first-pass priority such as SMOKE or REGRESSION.
Write structured outputs for downstream use.

Guardrails

No blind crawling.
No DOM-coupled selector strategy.
No infinite depth or uncontrolled exploration.
Every suggested test must map to a capability.

Output shape

{
  "capabilities": [
    {
      "name": "Authentication",
      "priority": "SMOKE",
      "tests": [
        "User can login with valid credentials",
        "User cannot login with invalid credentials"
      ]
    }
  ]
}

Commands

node agents/url-test-planner.js --url http://sut.testlab:3000

Expected artifacts

/test-results/agent-output/
  test-plan.json
  locator-map.json

Expected outputs

✔ Capability-oriented test plan derived from the target URL.
✔ Baseline smoke and regression candidates.
✔ Locator inventory aligned to user-facing structure.
✔ Stable JSON outputs for later agent stages.

Failure modes

Empty plan: page did not load correctly or extraction logic found no usable structure.
Duplicate capabilities: grouping logic is too weak or route naming is ambiguous.
Unstable output across runs: the discovery flow is nondeterministic and must be constrained.
Low-value test suggestions: capability mapping needs stronger business anchors.

Operational value

Establishes the front door for later agents.
Turns a URL into structured understanding instead of manual note-taking.
Feeds locator, triage, and release-gate systems later without changing the core contract.
Provides a concrete split between agent discovery work and human validation work.