Releases: webmaxru/agent-skills
Releases · webmaxru/agent-skills
v1.3.1
v1.3.0
language-detector-api
- Added multilingual input confidence distribution guidance
- Added OperationError for failed model initialization after download
prompt-api
- Aligned with spec-normative EXPERIMENTAL labels for params(), topK, and temperature (extension-only)
- Aligned with spec-normative DEPRECATED labels for measureInputUsage, inputUsage, inputQuota, onquotaoverflow
- Added Edge hardware requirements (Windows/macOS, storage, VRAM)
- Updated compatibility strategy and troubleshooting to reference spec authority over browser-specific docs
proofreader-api
- Added Edge Phi-4-mini model note for Proofreader API preview
- Added correctionExplanationLanguage guidance when includeCorrectionExplanations is true
translator-api
- Corrected Chrome privacy-masking behavior: all pairs report as downloadable until first creation
- Clarified Edge flag requirements: both translation-api and streaming-by-sentence flags required
webmcp
- Updated tool unregistration to use AbortSignal pattern (Chrome 148 transition)
- Added troubleshooting for pre-aborted AbortSignal silent registration skip
- Added reference note on silent skip behavior for already-aborted signals
webnn
- Expanded Worker context support to SharedWorker and ServiceWorker
- Added createConstantTensor() API and MLTensor constant attribute
- Added MLGraphBuilder constant overloads including tensor-based constant
- Clarified accelerated vs powerPreference priority and opSupportLimits rank ranges
writing-assistance-apis
- Corrected Chrome VRAM requirement from "more than 4 GB" to "at least 4 GB"
- Added Proofreader API cross-reference note in Chrome built-in AI family
- Added Edge "tl;dr" vs spec "tldr" discrepancy note for Summarizer type option
v1.2.1
v1.2.0
language-detector-api
- Added per-language availability checking via availability() starting with Chrome 132
- Documented spec-accurate filtering rule for omitting low-confidence results
- Added OperationError troubleshooting entry for failed model initialization
prompt-api
- Added Chrome hardware requirements (VRAM, RAM, storage thresholds)
- Documented regex responseConstraint support in both Chrome and Edge alongside JSON Schema
- Updated topK and temperature as browser-specific supported parameters in both browsers
- Added note about expired aiLanguageModelOriginTrial extension permission cleanup
proofreader-api
- Clarified ChromeOS support limited to Chromebook Plus devices from specific platform version
- Documented model auto-removal when storage drops below 10 GB in both Chrome and Edge
translator-api
- Clarified Edge sentence-level streaming flag behavior for improved responsiveness
- Documented offline translation capability after initial model download
- Improved NotReadableError description with example filtered content scenarios
webmcp
- Added AbortSignal option for registerTool() in Chrome 148+
- Documented unregisterTool() removal in Chrome 148 with migration guidance
- Updated template with AbortController-based tool lifetime management
- Clarified inputSchema as optional and documented UserInteractionCallback signature
webnn
- Removed deprecated MLDeviceType enum in favor of accelerated/powerPreference device options
- Updated API surface references and runtime template for current spec
writing-assistance-apis
- Documented option defaults for Summarizer, Writer, and Rewriter create() options
- Added preference semantics (auto, speed, capability) for Summarizer
- Documented model eviction when storage drops below 10 GB
v1.1.2
Full Changelog: v1.1.1...v1.1.2
v1.1.1
Full Changelog: v1.1.0...v1.1.1