Sap Sac Test Automation Agent Skill

SAP Analytics Cloud Test Automation

Design reusable SAC test automation as a capability-gated system: select the safest available discovery backend, require human review for profiles and baselines, then use reviewed Playwright code for deterministic execution, CI gating, reporting, and evidence.

Apply the core rule: discovery proposes, humans approve, Playwright executes, CI enforces.

Related Skills

• sap-sac-scripting: Use for Optimized Story Experience scripting, Analytics Designer APIs, diagnostic widgets, story script behavior, and SAC MCP setup.
• sap-sac-planning: Use for planning models, private/public versions, data actions, multi actions, validation rules, and writeback risk analysis.
• sap-sac-custom-widget: Use when dashboards include owned custom widgets that need stable ARIA labels, public test hooks, lifecycle fixes, or black-box test contracts.
• sap-dependency-security: Use before adding source-pinned browser tools, MCP servers, CI dependencies, or executable automation that handles SAC credentials.
• sap-browser-automation: Use for in-app manual authentication, consent-gated Edge profile reuse, fresh Edge/CDP startup, auth-state bootstrap, and browser recovery.
• agent-browser: Optionally load when the Vercel Labs agent-browser CLI is available and exact command syntax, snapshot/ref usage, screenshots, console, or network inspection is needed.
• playwright: Optionally load for CLI-based browser driving and debugging. For durable @playwright/test suites, use this SAC skill as the test architecture guide and follow the local project's Playwright conventions.
• chrome-devtools: Optionally load when Chrome DevTools MCP is installed and approved for read-only browser discovery, console/network inspection, screenshots, Lighthouse, or performance traces. Use references/chrome-devtools-mcp.md for SAC-safe defaults and Edge boundaries.

Initial Guidance

When the user is starting SAC automation or has not supplied a reviewed dashboard profile, route them through /sac-test-onboard or follow the same intake sequence manually. Use a two-stage intake:

• Stage 1: environment class, tenant risk, allowed discovery/execution tools, auth and roles, writeback/comment scope, and baseline ownership.
• Stage 2: planning/writeback, comments, permissions, visual/data baselines, custom widgets, CI stages, and failure evidence.

Default to draft-only artifacts until the user explicitly confirms file creation and target directory. If writing is confirmed, use profiles/<profile-id>/intake.md, profiles/<profile-id>/dashboard.yaml, and profiles/<profile-id>/scenarios/read-only-smoke.yaml based on the bundled templates.

After a profile or scenario draft exists, route safety review to sac-test-profile-reviewer when available. Keep the reviewer focused on intake/profile/scenario safety, not broad Playwright suite implementation.

When to Use This Skill

Use this skill to plan, implement, or review SAC automation involving:

• SAC stories, optimized stories, reports, dashboards, tabs, filters, prompts, tables, charts, exports, scripts, comments, or bookmarks.
• Dashboard profile files, scenario YAML, expected data files, visual baselines, Playwright adapters, and SAC-specific fixtures.
• Stored authentication state, SSO/MFA constraints, role-based test users, permission matrices, and CI gates.
• Planning writeback, private/public versions, data locks, validation rules, data actions, multi actions, or comment cleanup.
• Failure packets with traces, screenshots, videos, visual diffs, expected-vs-actual values, console/network summaries, and root-cause classification.

Do not use this skill as the main guide for generic web applications. Use general Playwright guidance for non-SAC sites.

Quick Start

• Classify the SAC story: production read-only, QA clone, planning/writeback, comments, permission-sensitive, or exploratory.
• Establish the automation policy: allowed tenants, allowed users, auth storage handling, writeback approval, baseline approval, and CI triggers.
• Run the capability and policy gate: choose manual discovery, Firecrawl public research, Chrome DevTools MCP with supported Chrome, Microsoft Edge/CDP, Chrome DevTools MCP with Edge best-effort, Playwright, agent-browser, or an approved enterprise browser lab based on what is installed and allowed.
• Run read-only discovery with the selected backend: capture snapshots, annotated screenshots, console/page errors, network clues, and candidate component maps without sending private SAC content to unapproved external services.
• Convert discovery into a human-reviewed dashboard profile: pages, widgets, locators, readiness markers, roles, data baselines, visual baselines, and known restrictions.
• Implement deterministic Playwright tests through reusable adapters. Keep scenario files selector-free; route interactions through component IDs from the profile.
• Gate only safe packs in CI first: auth check, read-only smoke, navigation, and critical widget readiness. Move planning, comments, data actions, visual baselines, and full permissions to controlled/nightly stages.
• Emit reviewer-friendly failure evidence: HTML report, trace, screenshots, video, visual diff, widget metadata, expected vs actual values, and failure category.

Read-Only Story Acceptance

Apply this contract to reporting-only SAC stories and dashboards. It complements the profile-driven smoke scenario and does not authorize planning or model changes.

Tenant and Model Safety

• Confirm the approved tenant, story, and data source/model before interaction.
• Verify that the current SAC area is Story Designer, not Modeler.
• Confirm that no model edit dialog, model save, master-data change, permission change, planning writeback, version publish, data action, multi action, or lock change occurred.
• Bind assertions only to dimensions, measures, hierarchies, filters, and comparison categories exposed by the approved model. If a required field is unavailable, omit the dependent widget and record the omission.

Story Readiness

• Open each page and wait for page-specific readiness markers, spinner disappearance, stable widget state, and the absence of visible SAC error messages.
• Verify that every KPI, chart, and table displays values or an explicit no-data state; loading completion alone is not sufficient.
• Confirm that filters populate, accept valid values, and update dependent widgets without changing the model.
• Exercise approved selections, navigation, and drilldowns where configured; verify that the resulting table or chart state is consistent.
• Record unresolved widgets, invalid bindings, unavailable fields, relevant console or network errors, and tenant-specific limitations.

Evidence

• Capture an approved screenshot or equivalent artifact for each story page and retain the story name and location.
• Pair visual evidence with KPI, table, or data assertions; do not use chart pixels as the only business-value check.
• Redact tenant identifiers, story IDs, session-like URL parameters, credentials, cookies, private screenshots, and unrelated tab content.

Browser Failure and Honest Handoff

Browser access is an execution prerequisite, not evidence that the story exists. If browser initialization or discovery fails, including runtime errors such as agent is not defined:

• Read the selected browser troubleshooting guidance and retry discovery without bypassing authentication or tenant permissions.
• If the normal browser path remains unavailable, use sap-browser-automation and the approved Microsoft Edge/CDP recovery guidance in references/edge-cdp-enterprise.md.
• If Edge/CDP is unavailable or blocked by policy, use approved desktop/manual assistance when available; ask the user to sign in interactively and never automate MFA.
• If authenticated tenant control still cannot be established, mark execution as blocked. Do not claim that the story was created or verified.
• Hand off an implementation-ready specification containing the story purpose and audience, confirmed model metadata, page-by-page widgets and filters, interactions, omitted or unresolved fields, read-only constraints, and the verification evidence still required.

Operating Model

Treat AI/browser-agent output as a draft, not as the source of truth. Require human review for profile creation, selector approval, expected business values, visual/data baseline changes, permission matrices, and any destructive/writeback scenario.

Prefer profile-driven automation:

• Store tenant/story metadata, pages, components, roles, baselines, and test policies in versioned YAML/JSON.
• Implement adapters for SAC component types such as buttons, tables, charts, tabs, filters, prompts, planning tables, comments, custom widgets, exports, and bookmarks.
• Keep selectors out of scenario files. Scenarios should call component IDs and actions, while adapters resolve locators and readiness behavior.
• Start with serial execution for each dashboard. Increase parallelism only after proving isolation for auth state, tenant state, comments, planning versions, live connections, and backend capacity.

Bundled Resources

Load these references only as needed:

• references/architecture.md: hybrid architecture, feasibility boundaries, reliable SAC test categories, and reusable project shape.
• references/tool-availability-and-deployment.md: backend decision matrix, Windows/restricted-environment checks, Firecrawl public-research policy, and no-tool fallbacks.
• references/chrome-devtools-mcp.md: Chrome DevTools MCP modes, SAC-safe configuration, tool categories, CLI usage, Windows/restricted deployment, and enterprise safety boundaries.
• references/edge-cdp-enterprise.md: SAC test-automation add-on for the shared sap-browser-automation authentication, profile-copy, Edge/CDP, and recovery layer.
• references/dashboard-profiles-and-scenarios.md: dashboard profile contract, scenario contract, adapter responsibilities, and onboarding flow.
• references/agent-browser-discovery.md: optional agent-browser read-only discovery workflow, command patterns, output artifacts, and human review checklist.
• references/playwright-execution.md: Playwright test runner guidance, auth, readiness, CI stages, and test category policy.
• references/governance-and-sac-testability.md: SAC testability contract, auth/SSO, planning/comment safety, baseline approval, and role governance.
• references/failure-triage-and-artifacts.md: required evidence, failure packet shape, root-cause categories, and performance/readiness metrics.
• templates/intake.md: guided intake packet for policy, tooling, roles, risk, baselines, and approvals.
• templates/dashboard-profile.yaml: starter dashboard profile with SAC metadata, readiness, components, roles, baselines, and risk policy.
• templates/scenario-read-only-smoke.yaml: selector-free starter smoke scenario using profile component IDs.

When implementing against a live project, also inspect the project's existing Playwright config, package manager, CI, profile schema, and artifact conventions before adding new structure.

Common Issues

• Avoid generated SAC DOM IDs and private framework classes. Prefer visible labels, ARIA roles, text, widget metadata, profile component IDs, and reviewed fallbacks.
• Do not treat story shell load as widget readiness. Wait for page markers, critical widgets, absence of SAC error text, spinner disappearance, stable values, and retrying assertions.
• Do not automate MFA as a release dependency. Prefer stored auth state, dedicated test users, security-approved test IdP policy, or manual refresh.
• Do not run comments, planning writeback, public version publish, data actions, or multi actions against production unless there is explicit formal approval.
• Do not approve visual or data drift automatically. Require owner review, baseline reason, and pull-request evidence.
• Do not use chart pixels as the only business-value assertion. Pair chart screenshots with table/KPI/data assertions where possible.
• Do not assume agent-browser, Playwright CLI, Chrome DevTools MCP, Firecrawl MCP, public npm, browser downloads, or remote debugging are available in company environments. Use the capability gate and document fallbacks.
• Do not use Chrome DevTools MCP as the audited CI release gate. Use it for discovery/debugging; convert approved findings into profile-driven Playwright tests.
• Do not run Chrome DevTools MCP against private SAC without disabling usage statistics, update checks, and CrUX field-data lookups, and without applying profile, URL, screenshot, and network-output controls.
• Do not expose CDP beyond loopback, publish webSocketDebuggerUrl, or bypass Edge RemoteDebuggingAllowed policy. Attaching to or copying a daily user profile requires explicit approval and the shared skill's isolated-profile procedure.
• Do not treat /json/version or /json/list returning 404 as proof that Edge CDP is unusable; read DevToolsActivePort and use the direct browser WebSocket only when the harness supports it.
• Do not send authenticated SAC tenant pages, screenshots, HARs, cookies, storage state, internal URLs, customer data, or private company docs to Firecrawl unless the exact deployment and retention mode are approved.
• Do not assume SAC optimized story features, tenant configuration, live data, localization, or prompt persistence behave identically across customers.

Source and Verification Notes

Derived from incorporated SAC automated-suite planning content recorded in docs/project/sac-test-automation-source-review-2026-06-17.md, plus extracted profile/scenario templates bundled with this skill. The planning sources cite SAP Help, Vercel Labs agent-browser, and Playwright documentation. Edge/CDP and Chrome DevTools MCP guidance also considers the ChromeDevTools/chrome-devtools-mcp README, CLI docs, tool reference, troubleshooting guide, package metadata, bundled skills, issue #1235 and PR #1229, Microsoft Edge DevTools Protocol documentation, Microsoft Edge DevTools MCP guidance, Edge RemoteDebuggingAllowed policy, and Firecrawl public documentation for MCP/search/scrape safety. This skill is docs-audited only; live SAC tenant execution, Chrome DevTools MCP runtime behavior, SSO behavior, CI behavior, planning writeback, and visual baseline stability remain tenant-specific and must be validated before making runtime claims.