Optional agent-browser Discovery Workflow

Source: Derived summary from incorporated SAC automated-suite planning content and local agent-browser skill behavior.

Load the agent-browser skill only when the CLI is installed and approved. Use tool-availability-and-deployment.md first in restricted environments, because agent-browser may be unavailable on managed Windows desktops, locked-down CI images, or networks that block tool installation.

Discovery Rules

Keep discovery read-only unless the user explicitly approves the action and target environment.
Use agent-browser output as a draft. Require human review before selectors, profiles, scenarios, baselines, or expected values become CI inputs.
Capture discovery artifacts in the dashboard profile directory, not as loose top-level files.
Prefer accessibility snapshots, annotated screenshots, visible labels, text, roles, and SAC widget names over private DOM selectors.
Re-run snapshots after tab changes, filter dialogs, prompts, overlays, or any UI state change.
If agent-browser is missing, use manual discovery, Chrome DevTools MCP with supported Chrome, Edge/CDP discovery, Chrome DevTools MCP with Edge best-effort, or Playwright exploration according to the capability gate.

Typical Commands

Use the local agent-browser skill for current CLI syntax. A typical SAC discovery session follows this shape only after the tool and data policy are approved:

agent-browser open "$SAC_STORY_URL"

agent-browser snapshot > profiles/finance-executive/discovery/agent-browser/snapshots/overview.txt

agent-browser screenshot \
  --annotate \
  --screenshot-dir profiles/finance-executive/discovery/agent-browser/screenshots

agent-browser console --json \
  > profiles/finance-executive/discovery/agent-browser/console.json

agent-browser errors \
  > profiles/finance-executive/discovery/agent-browser/page-errors.txt

agent-browser network har start

# Explore only approved read-only flows:
# - open tabs
# - inspect filters/prompts
# - inspect tables/charts
# - test read-only buttons/download affordances when allowed

agent-browser network har stop \
  profiles/finance-executive/discovery/agent-browser/network.har

If the installed CLI uses ref-based snapshots, follow the snapshot/ref loop: open, snapshot, interact with current refs, then snapshot again.

Discovery Output

Store:

profiles/<dashboard>/discovery/agent-browser/
  snapshots/
    overview.txt
    detail.txt
    planning.txt
  screenshots/
    overview-annotated.png
    detail-annotated.png
    planning-annotated.png
  proposed-profile.yaml
  proposed-scenarios.yaml
  console.json
  page-errors.txt
  network.har
  review-notes.md

Do not commit secrets, cookies, auth state, private browser profiles, or tenant-local credential files.

Human Review Checklist

Confirm:

Story ID, tenant/environment, open mode, bookmark, and optimized-story status.
Every page/tab, including overflow/role-hidden tabs.
Business-critical widgets and components to ignore.
SAC widget names from the story outline where possible.
Candidate selectors and any reviewed CSS/XPath fallbacks.
Page-ready markers and widget-specific readiness expectations.
Expected values, tolerances, data snapshot, and visual baseline stability.
Role expectations for visible, hidden, allowed, denied, and data-restricted behavior.
Planning reset strategy, private/public version policy, comment isolation, and cleanup.
Unsupported or restricted SAC features for the story/tenant.

Third-party checks

No alerts1mo3 checks · Risk SAFE

Gen Agent Trust Hub1mo
This skill provides a security-focused framework for automating SAP Analytics Cloud (SAC) testing using Playwright and Chrome DevTools. It identifies risks associated with browser automation in enterprise environments and implements strong mitigation strategies, including mandatory human review of all discovery artifacts, telemetry redaction, and the use of isolated browser profiles. The 'low' verdict reflects the inherent risk of an agent ingesting external web content during discovery, which is appropriately managed by the skill's design.
Socket1mo
No alerts
Snyk1mo
Risk: LOW · No issues

Capability

Other metadata

metadata: { "maintainer": "Eduard Jiglau", "maintainer_email": "hello@sap-ai-skills.com", "website": "https://sap-ai-skills.com", "version": "2.4.1", "last_verified": "2026-06-17", "sac_version": "2026.8", "documentation_source": "docs/project/sac-test-automation-source-review-2026-06-17.md", "primary_tools": [ "Playwright", "Chrome DevTools MCP", "Microsoft Edge CDP", "Vercel Labs agent-browser", "Manual discovery" ], "status": "docs_audited_runtime_pending", "known_issues": [ "Live SAC tenant, SSO, planning writeback, and CI execution checks require tenant-specific evidence before claiming runtime validation." ] }

/sap-sac-test-automation

referencesagent-browser-discovery.md