v1.10.1

• Updated toolactivated and toolcancel guidance for their move to document.modelContext in Chrome 156.0.8076.0.
• Added a feature-detected listener example that remains compatible with older Chrome builds using window.
• Updated compatibility, API reference, and troubleshooting guidance for the new event target.

v1.10.0

• Added annotations.debugging guidance for Chrome 156+ so inspection and developer-focused tools can be identified and filtered by general-purpose agents
• Added debugging to the TypeScript tool-registration template
• Added compatibility details and a diagnostic-tool registration example based on WebMCP PR #253

v1.9.0

language-detector-api

• Corrected the cumulative-confidence cutoff: languages are omitted once confidence reaches or exceeds 0.99.

webmcp

• Added Chrome 154 consequentialHint guidance for high-stakes, irreversible, and real-world actions.
• Added independent confirmation and authorization requirements plus validation guidance.
• Added typed template support and a consequential flight-booking example.

v1.8.0

language-detector-api

• Clarified Chrome's desktop-only support and removed stale, duplicated compatibility notes
• Documented Microsoft Edge 148+ as authoritative despite lagging compatibility tables
• Added cross-browser downloadprogress handling for fractional and byte-based progress values

prompt-api

• Updated guidance for Chrome 148 shipping support and Edge's Aion CPU-capable model path
• Marked tool calling and samplingMode as experimental, and sampling parameters as deprecated for portable pages
• Updated templates and references for required context-usage APIs with older-browser fallbacks
• Added QuotaExceededError prevention and recovery guidance
• Tightened availability and session guidance around experimental options

proofreader-api

• Updated Chrome local testing to use the dedicated #proofreader-api flag

webmcp

• Added Chrome 153 per-execution cancellation through the execute(input, { signal }) contract
• Separated registration lifecycle cleanup from in-flight execution cancellation
• Added declarative executeTool() cancellation guidance and a working example
• Replaced obsolete requestUserInteraction() guidance with visible application UI and authorization checks
• Expanded troubleshooting and documented current polyfill cancellation limitations
• Updated templates and target discovery for the current imperative API surface

webnn

• Clarified when SharedWorker and ServiceWorker execution contexts became supported
• Refined opSupportLimits() fingerprinting guidance and future MLContextOptions considerations

writing-assistance-apis

• Updated Edge flags, device-performance guidance, and Aion CPU-inference support
• Added NotReadableError handling for browser-filtered output

v1.7.1

webmcp

• Register WebMCP tools concurrently with Promise.all in the model-context-registry.template.ts asset, so total registration time is bounded by the slowest tool instead of the sum of all registrations.
• Kept a per-tool try/catch so one failing registration never blocks the others, and preserved the async-aware ready promise and synchronous dispose() contract.

v1.7.0

language-detector-api

• Updated Microsoft Edge support: the Language Detector API is supported from Edge 148 with no flags required (previously documented as a flagged developer preview in 147).

webmcp

• Documented that document.modelContext.registerTool() returns a Promise as of Chrome 151.0.7922.0, resolving only when the tool is available to getTools() across the frame tree.
• Updated the registration guidance to await registerTool() inside try/catch, backward-compatible with older synchronous Chrome builds.
• Made the model-context-registry.template.ts asset async-aware: registration awaits inside try/catch, exposes a ready promise, and keeps dispose() synchronous.
• Added troubleshooting guidance for uncaught registerTool() failures on Chrome 151+.
• Expanded compatibility and reference notes for the new asynchronous registration semantics.

webnn

• Documented context.lost as Promise<MLContextLostInfo> carrying a human-readable message field.
• Added MLOperand/MLTensor dataType and shape readonly attributes (property access, not method calls).
• Clarified the builder.constant(dataType, value) scalar overload using MLNumber.
• Added a note that MLContextOptions is evolving (fallback and device-preference controls) so detection and fallback paths stay forward-compatible.
• Recommended optional operator label properties for more diagnosable build()/dispatch() error messages.

v1.6.0

• Migrated webmcp tool registration to document.modelContext with a navigator.modelContext fallback for Chrome 146–149
• Documented Chrome 150 deprecation of navigator.modelContext in webmcp compatibility and reference notes
• Updated the webmcp registry template to feature-detect the model context across both surfaces
• Refreshed webmcp troubleshooting guidance for the undefined modelContext case
• Added a document.modelContext marker to the webmcp find-webmcp-targets scanner
• Fixed repository URLs across plugin configs and README to point to web-ai-agent-skills

v1.5.0

webmcp

• Added support for the new tool name constraint (1-128 ASCII alphanumeric, _, -, .) across SKILL, reference, and troubleshooting docs
• Added title field and untrustedContentHint annotation to the tool definition contract and template
• Documented the new toolchange event on ModelContext and the exposedTo registration option
• Documented NotAllowedError (Permissions Policy) and SecurityError (non-trustworthy exposedTo origins) handling

proofreader-api

• Corrected Chrome minimum VRAM requirement to strictly more than 4 GB in the compatibility reference

Repository maintenance

• Fixed installation command for web-ai-skills plugin in README
• Corrected Claude Code plugin install identifier in README
• Pinned Copilot CLI to 1.0.43 in the skill-update-worker agentic workflow
• Added version-bump-and-deploy maintenance prompt to APM dependencies

v1.4.0

language-detector-api

• Added Chrome desktop-only restriction to compatibility notes
• Clarified expectedInputLanguages nullability
• Documented Chrome mobile limitation and measureInputUsage/Infinity behavior

prompt-api

• Added tool-call and tool-response content types from the April 2026 spec
• Documented Chrome audio-input GPU requirement
• Updated language-model-service template with new content types

proofreader-api

• Corrected Chrome VRAM requirement from >4 GB to ≥4 GB

translator-api

• Added desktop-only constraint, storage, download-failure, and inputQuota notes
• Expanded translator reference and troubleshooting guidance

webnn

• Added Origin Trial note and opSupportLimits fingerprinting caveat in compatibility/reference

writing-assistance-apis

• Corrected Chrome VRAM requirement and added @types/dom-chromium-ai guidance

v1.3.1

• Bumped internal skill metadata version from 1.1 to 1.2 for all skills: language-detector-api, prompt-api, proofreader-api, translator-api, webmcp, webnn, writing-assistance-apis