Assessing External Test Risk Agent Skill

Assessing external test risk

Use this skill to decide whether a branch or PR should include external e2e coverage using @pytest.mark.external_test.

This helps protect deployments that commonly involve proxies, embedded iframe contexts, CSP constraints, and other browser security boundaries.

This skill is for risk assessment and recommendation. It does not auto-mark tests unless explicitly requested.

Decision rule

Use an any-hit policy:

• If any checklist category is hit, output Recommend external_test: Yes
• If no categories are hit, output Recommend external_test: No

Inputs to review

• Branch or PR diff against its base branch
• Changed files and related tests
• PR description (if available)

Assessment workflow

• Gather the changed files and full diff against the base branch.
• Evaluate each checklist category below as hit or not hit.
• Record concrete evidence from file paths and diff snippets.
• Produce a recommendation and specific external-test focus areas.

Checklist categories

Evaluate all categories. A single hit is enough to recommend external coverage.

• Routing and URL behavior
• Hit when changes introduce or modify Starlette routes, server.baseUrlPath, catch-alls, request methods, URL resolution, redirects, or status codes.

• Auth, cookies, CSRF, and identity binding
• Hit when changes touch login/logout or OAuth flows, _streamlit_user, _streamlit_xsrf, CSRF/XSRF handling, server.trustedUserHeaders, or session-to-identity binding.

• Websocket handshake and session transport
• Hit when changes affect websocket handshake or subprotocols, session affinity, reconnect behavior, ping or timeout behavior, message size limits, or fragmentation.

• Embedding and iframe boundary
• Hit when changes modify host-to-guest communication (postMessage), iframe sizing or resize behavior, iframe sandbox or allow attributes, or permissions policy behavior in embedded contexts.

• Static and component asset serving
• Hit when changes alter asset handlers, cache headers, size limits, base paths (including server.customComponentBaseUrlPath), or proxying rules for static/component assets.

• Service worker, uploads, and downloads
• Hit when changes modify service worker registration, scope, or caching strategy; upload/download endpoints; JWT or CSRF wrapping; or download attribute behavior.

• Cross-origin behavior and external networking
• Hit when changes alter CORS allowlists, crossOrigin usage, external-origin fetches or external networks behavior, or backend URL discovery via window.__streamlit.*.

• Cross-origin theming and resource discovery
• Hit when changes introduce or modify theme/resource loading across origins (fonts, images, theme globals), CSS isolation with host pages, or manifest/asset discovery when HTML is not served by Starlette.

• SiS and Snowflake runtime dependencies
• Hit when changes rely on or modify SiS/Snowflake runtime behavior, including running_in_sis(), get_active_session(), Snowflake connection/session semantics, or SiS-specific environment flags.

10. Client storage behavior

• Hit when changes introduce or modify cookies, localStorage, or sessionStorage usage that may differ in embedded or third-party contexts.

11. Security headers and browser policies

• Hit when changes adjust CSP, Referrer-Policy, Permissions-Policy, or related headers that can impact embedding or resource loading.

Output format

Use this exact structure:

## External test recommendation

- Recommend external_test: [Yes/No]
- Triggered categories: [List category numbers and names, or "None"]
- Evidence:
  - `<path>`: [short reason from diff]
  - `<path>`: [short reason from diff]
- Suggested external_test focus areas:
  - [Concrete scenario to validate externally]
  - [Concrete scenario to validate externally]
- Confidence: [High/Medium/Low]
- Assumptions and gaps: [Unknowns, missing context, or why confidence is reduced]

Interpretation guidance

• Prefer evidence over intuition. Tie each hit to concrete diff details.
• When in doubt, err toward Yes if externally hosted or embedded behavior could diverge from local runs.
• Keep focus areas specific and testable (route, auth handshake, iframe boundary, asset loading, SiS runtime behavior).

Examples

Example yes recommendation

Diff includes:

• lib/streamlit/web/server/starlette/starlette_routes.py route changes
• Cookie/XSRF handling updates in request auth middleware
• Frontend embed code changing iframe allow attributes

Expected output:

• Recommend external_test: Yes
• Triggered categories include routing, auth/cookies/CSRF, and embedding boundary
• Focus areas include external host iframe embedding + auth/session continuity checks

Example no recommendation

Diff includes:

• Pure refactor in internal utility functions with no network, auth, embedding, storage, or runtime integration impact
• Docs and test name cleanup only

Expected output:

• Recommend external_test: No
• Triggered categories: None
• Confidence is high if no indirect integration points are touched