v4.22.0

Minor Changes

• ee1753d: Send adcp_major_version on every request per adcontextprotocol/adcp#1959. Sellers can validate the declared version against their supported range and return VERSION_UNSUPPORTED on mismatch.
• 68da21e: Add serve() helper for zero-boilerplate agent HTTP servers, fix examples to use npm-consumer import paths, and ship docs/llms.txt + BUILD-AN-AGENT.md in the npm package for agent discoverability.
• ea93508: Migrate comply() to storyboard-driven testing. The compliance engine now runs storyboard YAMLs instead of hand-written scenario functions. Adds YAML format extensions (expect_error, requires_tool, context_outputs/context_inputs, error_code validation) and 10 new compliance storyboards covering governance, SI, brand rights, state machines, error compliance, schema validation, behavioral analysis, audiences, and deterministic testing. Deprecates SCENARIO_REQUIREMENTS, DEFAULT_SCENARIOS, and testAllScenarios() in favor of storyboard execution.
• ea93508: Add storyboard-driven testing module with CLI support. Storyboards are YAML-defined test workflows that map directly to SingleAgentClient methods, enabling step-by-step agent testing. Includes 12 bundled storyboards from the AdCP spec, a stateless per-step CLI (adcp storyboard step) designed for LLM consumption, and platform type tags for backwards compatibility with the existing compliance system.
• e5002a4: Add userAgent config to PropertyCrawlerConfig and TestOptions, threaded through to all outbound HTTP requests via both MCP and A2A transports. Wire the existing but unused SingleAgentClientConfig.userAgent field into protocol headers. Export PropertyCrawlerConfig type from public API.

Patch Changes

• 913fadd: Add generated agent documentation (llms.txt, TYPE-SUMMARY.md) and update SKILL.md with all 24 test scenarios
• cc07055: Fix skipped-step counting in storyboard runner and add tool_discovery diagnostic observations to comply(). Steps skipped due to requires_tool are now correctly counted as skipped instead of passed, and comply() emits observations showing discovered tools and expected-vs-actual tools when tracks are skipped.