[anycodes]

Description

A developer following maintained guides could install an obsolete Git snapshot, pass decorator/MCP keywords absent from current signatures, or configure MCP with credentials it does not read. Use the published 0.4.0a2 pair, canonical sandbox_* arguments and explicit ProviderSandbox imports; keep genuine standard-client and third-party contracts distinct. Refs #111.

Type of Change

• Documentation

Changes

• Replace old installation pins and current first-party examples; preserve labeled historical records.
• Rewrite the beginner tutorial around usable base command/file transfers with owned cleanup.
• Use unified template launch and explicit application-token binding in the Express guide.
• Remove retired Python gRPC engine instructions from the maintained Volume guide.
• Retain valid native synchronous wrappers; no new aliases or runtime changes.

Testing

• Fresh official public a2 install, pip check and public imports.
• Extract and run complete EN/ZH beginner scripts against real Express: create, execute, write/read/download/upload, kill; both independent typed absence checks PASS.
• Public decorator/MCP signatures verified; documented MCP HTTP app health200/auth401 with exact public [mcp] installation PASS.
• Rewritten Python fences compile; repository health106/1warning/0fail, hygiene and diff checks PASS.

Checklist / Definition of Done

• Bilingual user docs, AGENTS entry, owner note/evidence and changelog.
• No credentials, no owned sandbox leaks, no product runtime change.
• Latest-head required CI and independent review before merge.

Broader public example acceptance remains PARTIAL:13/14 cloud examples passed; browser cold-install timeout remains recorded and its template policy awaits user choice. This PR does not close Issue111 or claim the template matrix passed. Details: .agents/evidence/verify/2026-10-25-public-sdk-documentation.md.

Fixes #118: creation benchmark now uses the installed public ProviderSandbox API, canonical fields, three fixed samples and explicit per-instance cleanup. Three failure/lifecycle regressions and an actual public-a2 run with three independent typed absence checks pass. Public install instructions explicitly select the official index; release guide states current published a2 facts.