@framers/agentos-skills-registry 0.8.0 → 0.9.0

package/README.md

@@ -6,7 +6,7 @@
 
 # @framers/agentos-skills-registry
 
-Curated skills registry for [AgentOS](https://github.com/framersai/agentos) — 40 SKILL.md prompt modules, typed catalog, and lazy-loading factories.
+Curated catalog of 60+ AgentOS skills with query helpers and lazy-loading factories.
 
 [![npm](https://img.shields.io/npm/v/@framers/agentos-skills-registry?logo=npm&color=cb3837)](https://www.npmjs.com/package/@framers/agentos-skills-registry)
 
@@ -14,15 +14,36 @@ Curated skills registry for [AgentOS](https://github.com/framersai/agentos) —
 npm install @framers/agentos-skills-registry
 ```
 
-## What's Inside
+## What This Package Is
 
-This is the **single package** for AgentOS skills. It contains:
+This is the **skills catalog** — the data layer that ships 60+ curated SKILL.md
+prompt modules and provides typed query helpers, search functions, and lazy-loading
+factories for consuming them.
 
-- **40 curated SKILL.md files** — prompt modules spanning social automation, developer tooling, productivity, research, voice, and more
-- **registry.json** — machine-readable index of all skills with metadata
-- **Static catalog** (`SKILLS_CATALOG`) — typed array with query helpers
-- **Registry factories** — `createCuratedSkillRegistry()`, `createCuratedSkillSnapshot()` (requires `@framers/agentos`)
-- **Validation script** — `npm run validate` to lint SKILL.md files
+It is **not** the skills engine. The engine lives in
+[`@framers/agentos-skills`](https://www.npmjs.com/package/@framers/agentos)
+(exported from `@framers/agentos/skills`), which provides the runtime
+`SkillRegistry`, `SkillSnapshot` builder, frontmatter parser, and eligibility
+resolver.
+
+### Architecture: catalog vs. engine
+
+```
+@framers/agentos              ← the skills ENGINE (runtime)
+  └── /skills                    SkillRegistry, SkillSnapshot, parser, eligibility
+        ▲
+        │  lazy import()
+        │
+@framers/agentos-skills-registry   ← THIS package (CATALOG)
+  ├── registry/curated/*/SKILL.md  60+ bundled prompt modules
+  ├── registry.json                machine-readable index of all skills
+  ├── catalog.ts                   SKILLS_CATALOG array + query helpers (zero deps)
+  └── index.ts                     factory functions that lazy-import the engine
+```
+
+**Dependency direction:** this catalog package depends on `@framers/agentos`
+(optional peer dep), never the other way around. The engine knows nothing about
+the catalog — it just provides the parsing and registry machinery.
 
 ## Quick Start
 
@@ -67,7 +88,7 @@ Access the JSON index directly:
 import { getSkillsCatalog } from '@framers/agentos-skills-registry';
 
 const catalog = await getSkillsCatalog();
-console.log(catalog.skills.curated.length); // 40
+console.log(catalog.skills.curated.length); // 60+
 console.log(catalog.version); // '1.0.0'
 ```
 
@@ -80,7 +101,8 @@ console.log(registry.skills.curated[0].name); // 'weather'
 
 ### 3. Dynamically load skills into an agent (requires @framers/agentos)
 
-The factory functions lazy-load `@framers/agentos` via dynamic `import()`:
+The factory functions lazy-import `@framers/agentos/skills` (the engine) via
+dynamic `import()` — resolved only when you call them, cached after first use:
 
 ```bash
 npm install @framers/agentos-skills-registry @framers/agentos
@@ -149,23 +171,26 @@ const valid = skillNames.filter((name) => {
 const snapshot = await createCuratedSkillSnapshot({ skills: valid });
 ```
 
-When `skills` is a string array, the registry only loads those specific `SKILL.md`
+When `skills` is a string array, the catalog only loads those specific `SKILL.md`
 files before building the snapshot. It does not walk the full curated bundle first.
 Loaded skills also include parsed `metadata` so consumers do not need to decode
 the `metadata.agentos` block manually.
 
-## Two Import Paths
+## Sub-exports
 
-| Import | Peer deps | Use case |
-|--------|-----------|----------|
-| `@framers/agentos-skills-registry/catalog` | None | UI browsing, search, filtering |
-| `@framers/agentos-skills-registry` | `@framers/agentos` (optional) | Runtime loading, snapshots, factories |
+| Export path | Peer deps | Use case |
+|-------------|-----------|----------|
+| `@framers/agentos-skills-registry` | `@framers/agentos` (optional) | Full SDK: catalog + factory functions + schema types |
+| `@framers/agentos-skills-registry/catalog` | None | Lightweight: `SKILLS_CATALOG`, query helpers (search, filter, browse) |
+| `@framers/agentos-skills-registry/registry.json` | None | Raw JSON index of all skills |
+| `@framers/agentos-skills-registry/workspace-discovery` | None | Discover SKILL.md files in workspace directories |
+| `@framers/agentos-skills-registry/types` | None | TypeScript declarations for registry.json schema |
 
-The `@framers/agentos` dependency is loaded **lazily** at runtime and cached after first resolution. If it's not installed and you call a factory function, you get a clear error with install instructions.
+The `@framers/agentos` dependency is loaded **lazily** at runtime and cached after first resolution. If it is not installed and you call a factory function, you get a clear error with install instructions. The catalog query helpers work without it.
 
-## Included Skills (40)
+## Included Skills (60+)
 
-The catalog now includes both foundational utility skills and social automation modules, including:
+The catalog includes both foundational utility skills and social automation modules:
 
 - Information and research: `web-search`, `weather`, `summarize`, `deep-research`
 - Developer tools: `github`, `coding-agent`, `git`
@@ -203,26 +228,6 @@ import type {
 // SkillInstallSpec — install instructions for skill dependencies
 ```
 
-## Exports
-
-| Export path | Contents |
-|-------------|----------|
-| `.` | Full SDK: catalog helpers + factory functions + schema types |
-| `./catalog` | Lightweight: `SKILLS_CATALOG`, query helpers (zero peer deps) |
-| `./registry.json` | Raw JSON index of all skills |
-| `./types` | TypeScript declarations for registry.json schema |
-
-## Relationship to Other Packages
-
-```
-@framers/agentos-skills-registry     ← This package (data + SDK)
-  ├── registry/curated/*/SKILL.md    (bundled prompt modules)
-  ├── registry.json                  (machine-readable index)
-  ├── catalog.ts                     (typed queries: search, filter, browse)
-  └── index.ts                       (factories: lazy-load @framers/agentos)
-        └── @framers/agentos         (optional peer: live SkillRegistry + snapshots)
-```
-
 ## Contributing
 
 See [CONTRIBUTING.md](./CONTRIBUTING.md) for how to submit new skills.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@framers/agentos-skills-registry",
-  "version": "0.8.0",
+  "version": "0.9.0",
   "files": [
     "dist",
     "registry",
@@ -10,7 +10,7 @@
     "CONTRIBUTING.md",
     "README.md"
   ],
-  "description": "Curated skills registry for AgentOS — SKILL.md prompt modules, typed catalog, and lazy-loading factories",
+  "description": "Curated skills catalog for AgentOS — 60+ bundled SKILL.md prompt modules with query helpers and lazy-loading factories",
   "type": "module",
   "sideEffects": false,
   "main": "./dist/index.js",

package/registry/curated/agent-config/SKILL.md

@@ -0,0 +1,22 @@
+---
+name: agent-config
+description: Export and import agent configurations for sharing and backup
+version: 1.0.0
+tags: [agent, config, export, import, yaml, json]
+tools_required: []
+---
+
+# Agent Configuration
+
+Export agent configurations as portable YAML/JSON files for sharing, backup, and migration. Import configurations to recreate agents.
+
+## Capabilities
+- **Export**: Save agent config with instructions, tools, personality, guardrails
+- **Import**: Recreate agent from exported config
+- **Secret redaction**: API keys automatically redacted on export
+- **Round-trip**: Export -> edit -> import workflow
+
+## Example
+"Export my research agent to a file"
+"Import this agent configuration"
+"Share my agent config with the team"

package/registry/curated/amazon-polly/SKILL.md

@@ -0,0 +1,74 @@
+---
+name: amazon-polly
+version: '1.0.0'
+description: Neural text-to-speech via Amazon Polly — high-quality voices, MP3 output, voice listing, default Joanna (en-US Neural).
+author: Wunderland
+namespace: wunderland
+category: voice
+tags: [voice, tts, text-to-speech, amazon, aws, polly, neural]
+requires_secrets: [aws.accessKeyId, aws.secretAccessKey]
+requires_tools: []
+metadata:
+  agentos:
+    emoji: "\U0001F50A"
+    primaryEnv: AWS_ACCESS_KEY_ID
+    homepage: https://docs.aws.amazon.com/polly/
+---
+
+# Amazon Polly TTS
+
+Use this skill when the agent is deployed on AWS infrastructure or when the user prefers Amazon Polly's Neural engine voices. Polly provides high-quality neural TTS with MP3 output and a wide range of voices across languages and regions.
+
+Prefer this over OpenAI or ElevenLabs TTS when the user already has AWS credentials configured, or when cost efficiency at high volume is a priority (Polly pricing is per-character rather than per-request).
+
+## Setup
+
+Set the following in the environment or agent secrets store:
+
+| Variable                  | Description              |
+|---------------------------|--------------------------|
+| `AWS_ACCESS_KEY_ID`       | IAM access key ID        |
+| `AWS_SECRET_ACCESS_KEY`   | IAM secret access key    |
+| `AWS_REGION`              | AWS region (default `us-east-1`) |
+
+## Configuration
+
+```json
+{
+  "voice": {
+    "tts": "amazon-polly"
+  }
+}
+```
+
+With a specific voice:
+
+```json
+{
+  "voice": {
+    "tts": "amazon-polly",
+    "providerOptions": {
+      "voice": "Matthew"
+    }
+  }
+}
+```
+
+## Provider Rules
+
+- Default voice is `Joanna` (en-US, Neural engine).
+- Always use the Neural engine for production — Standard voices are lower quality.
+- Call `listAvailableVoices()` to enumerate all voices available in the configured region.
+- Audio is returned as MP3. Playback requires an MP3-capable audio pipeline.
+
+## Examples
+
+- "Use Amazon Polly with the Matthew voice for this agent."
+- "List available Polly voices in us-east-1."
+- "Synthesize this message using Amazon Polly."
+
+## Constraints
+
+- Requires `AWS_ACCESS_KEY_ID` and `AWS_SECRET_ACCESS_KEY` with `polly:SynthesizeSpeech` and `polly:DescribeVoices` IAM permissions.
+- Neural engine voices are region-specific. Not all voices are available in every AWS region.
+- API costs apply per character synthesized (Neural engine pricing).

package/registry/curated/diarization/SKILL.md

@@ -0,0 +1,83 @@
+---
+name: diarization
+version: '1.0.0'
+description: Speaker diarization — identifies and tracks who is speaking at each moment in an audio stream, using provider-delegated labels or local offline clustering.
+author: Wunderland
+namespace: wunderland
+category: voice
+tags: [voice, diarization, speaker-identification, multi-speaker, offline, deepgram]
+requires_secrets: []
+requires_tools: []
+metadata:
+  agentos:
+    emoji: "\U0001F465"
+    homepage: https://docs.wunderland.sh/guides/voice
+---
+
+# Speaker Diarization
+
+Use this skill when the agent needs to track who is speaking across a multi-speaker audio stream. Supports two modes: provider-delegated (extracts speaker labels from Deepgram word-level results) and local clustering (spectral-centroid agglomerative clustering, fully offline).
+
+Enable this when transcribing meetings, interviews, podcast recordings, or any scenario where knowing which speaker said what is important for downstream tasks (summaries, action items, CRM updates).
+
+## Setup
+
+No API key required for local mode. For provider-delegated mode, enable diarization on the STT provider:
+
+```json
+{
+  "voice": {
+    "stt": "deepgram",
+    "diarization": "provider",
+    "providerOptions": { "diarize": true }
+  }
+}
+```
+
+For fully offline local clustering:
+
+```json
+{
+  "voice": {
+    "diarization": "local"
+  }
+}
+```
+
+## Speaker Enrollment
+
+Pre-register known speakers so the engine labels them by name instead of `Speaker_0`, `Speaker_1`:
+
+```ts
+await session.enrollSpeaker('Alice', aliceVoiceprintFloat32Array);
+await session.enrollSpeaker('Bob', bobVoiceprintFloat32Array);
+```
+
+## Provider Rules
+
+- Prefer provider-delegated mode with Deepgram when speaker accuracy is critical and `DEEPGRAM_API_KEY` is available. Word-level speaker labels are more reliable than local clustering.
+- Use local mode when privacy, offline operation, or zero additional API cost is required.
+- The local backend extracts 16-dimensional spectral feature vectors per 1.5 s window (0.5 s overlap) — suitable for clear audio. Replace with an ONNX x-vector model for noisy environments.
+- Enroll known speakers when participant names are known in advance to get named labels instead of generic `Speaker_N` identifiers.
+
+## Events
+
+| Event                 | Description                                              |
+|-----------------------|----------------------------------------------------------|
+| `speaker_identified`  | Active speaker label has changed                         |
+| `segment_ready`       | A labelled audio or transcript segment is ready          |
+| `error`               | Unrecoverable diarization error                          |
+| `close`               | Session fully terminated                                 |
+
+## Examples
+
+- "Transcribe this meeting and label each speaker's turns."
+- "Use local diarization for a private interview recording."
+- "Enable Deepgram diarization and label Alice and Bob by voice."
+
+## Constraints
+
+- Local mode accuracy depends on audio clarity and spectral separation between speakers.
+- Provider mode requires `diarize: true` support on the active STT provider (currently Deepgram).
+- Speaker enrollment voiceprints must be Float32Arrays computed from clean reference audio.
+- The built-in feature extractor is intentionally lightweight; replace `extractSimpleEmbedding()` with an ONNX x-vector model for production-quality voiceprints.

package/registry/curated/emergent-tools/SKILL.md

@@ -0,0 +1,97 @@
+---
+name: emergent-tools
+version: '1.0.0'
+description: Forge new tools at runtime via LLM — sandboxed execution, LLM-as-judge validation, composable tool building, and full audit trail.
+author: Wunderland
+namespace: wunderland
+category: productivity
+tags: [emergent, tools, forge, sandbox, dynamic, runtime, LLM-judge]
+requires_secrets: []
+requires_tools: [forge_tool]
+metadata:
+  agentos:
+    emoji: "\U0001F527"
+---
+
+# Emergent Tools
+
+You have access to the EmergentCapabilityEngine — a system that lets you create brand-new tools at runtime when no existing tool satisfies the user's request. This is a powerful capability; use it wisely.
+
+## When to Forge vs. Use Existing Tools
+
+Before forging a new tool, always check whether an existing tool can fulfill the request:
+
+1. **Search first** — Use `discover_capabilities` to scan the tool registry. If a tool already exists that handles the task (even partially), prefer it.
+2. **Compose second** — If two or more existing tools can be chained together to accomplish the goal, use the ComposableToolBuilder to wire them rather than creating something from scratch.
+3. **Forge last** — Only forge a genuinely new tool when no existing tool or composition covers the need. Common forge-worthy scenarios:
+   - A domain-specific data transformation not covered by general utilities
+   - A custom API integration the user needs on the fly
+   - A specialized validation or formatting pipeline
+   - A one-off computation that would be awkward to express as a prompt
+
+## The Forging Process
+
+When you decide to forge a tool, the pipeline works as follows:
+
+1. **Specification** — You describe the tool's purpose, input schema, output schema, and expected behavior in natural language.
+2. **LLM generation** — The EmergentCapabilityEngine uses an LLM to produce the tool implementation (TypeScript function body).
+3. **Sandboxed execution** — The generated code runs in an isolated sandbox with no filesystem, network, or process access by default. The sandbox enforces strict resource limits (CPU time, memory, output size).
+4. **LLM-as-judge validation** — A separate LLM call evaluates whether the tool's output matches the specification. The judge scores correctness, safety, and completeness.
+5. **Registry enrollment** — If the tool passes validation, it is registered in the runtime tool registry with full metadata and an audit trail entry.
+
+## Using ForgeToolMetaTool
+
+The `forge_tool` meta-tool is your interface to the EmergentCapabilityEngine. Invoke it with:
+
+- **name** — A clear, snake_case identifier for the new tool (e.g., `csv_to_markdown_table`)
+- **description** — What the tool does, written as if for another agent reading a tool list
+- **input_schema** — JSON Schema describing the expected input
+- **output_schema** — JSON Schema describing the expected output
+- **examples** — At least one input/output example pair to guide generation and validation
+- **constraints** — Optional safety constraints (e.g., "must not make network calls", "output must be valid JSON")
+
+The more precise your specification, the higher the first-pass success rate.
+
+## ComposableToolBuilder
+
+For compositions of existing tools, use the ComposableToolBuilder pattern:
+
+- **pipeline(tools[])** — Chain tools sequentially, piping each output as the next input
+- **parallel(tools[])** — Run tools concurrently and merge their outputs
+- **conditional(predicate, ifTool, elseTool)** — Branch based on a runtime condition
+- **transform(tool, mapFn)** — Wrap a tool with an output transformation
+
+Composed tools are registered just like forged tools, with full provenance tracking showing which base tools were combined.
+
+## EmergentJudge Quality Thresholds
+
+The LLM-as-judge system uses three thresholds:
+
+- **Correctness** (>= 0.8) — Does the output match the specification and examples?
+- **Safety** (>= 0.9) — Does the tool avoid side effects, data leaks, or dangerous operations?
+- **Completeness** (>= 0.7) — Does the tool handle edge cases and produce well-structured output?
+
+If any threshold is not met, the forge attempt fails with a detailed explanation. You can revise the specification and retry. Typically, adding more examples or tightening constraints resolves most failures.
+
+## Audit Trail
+
+Every forged tool carries an audit record containing:
+
+- The original specification
+- The generated source code (hash-pinned)
+- Judge scores and rationale
+- Timestamp and session context
+- Parent tool references (for compositions)
+
+This trail is immutable. If a user asks "how was this tool made?", you can retrieve and explain its provenance.
+
+## Best Practices
+
+1. **Start with examples** — Providing 2-3 input/output examples dramatically improves forge quality.
+2. **Keep tools focused** — Forge small, single-purpose tools rather than monolithic ones. Compose them later if needed.
+3. **Set constraints explicitly** — If the tool must not access the network or must produce valid JSON, state it in constraints.
+4. **Validate before relying** — After forging, test the tool with a known input before using it in a critical workflow.
+5. **Reuse forged tools** — Forged tools persist in the session registry. Check before forging a duplicate.
+6. **Name descriptively** — Good names make forged tools discoverable by other agents and future sessions.
+7. **Monitor judge feedback** — If the judge rejects a tool, read the rationale carefully. It usually pinpoints exactly what to fix.
+8. **Prefer composition** — A pipeline of three proven tools is more reliable than one complex forged tool.

package/registry/curated/endpoint-semantic/SKILL.md

@@ -0,0 +1,72 @@
+---
+name: endpoint-semantic
+version: '1.0.0'
+description: Semantic endpoint detection — uses an LLM to classify whether the user's utterance is a complete thought, reducing false turn boundaries on mid-sentence pauses.
+author: Wunderland
+namespace: wunderland
+category: voice
+tags: [voice, endpointing, turn-detection, semantic, llm, vad, silence]
+requires_secrets: []
+requires_tools: []
+metadata:
+  agentos:
+    emoji: "\U0001F4AC"
+    homepage: https://docs.wunderland.sh/guides/voice
+---
+
+# Semantic Endpoint Detector
+
+Use this skill when the agent's default silence-based turn detection is causing false positives — triggering agent responses mid-sentence when the user pauses to think. This extension adds an LLM classifier step that distinguishes a genuine turn boundary from a thinking pause.
+
+Prefer this over pure VAD/silence endpointing in conversational contexts where users speak with frequent mid-thought pauses, or when the user repeatedly complains that the agent "interrupts" them.
+
+## How It Works
+
+1. If the final transcript ends with `.`, `?`, or `!`, the turn ends immediately (punctuation path).
+2. Short acknowledgement phrases (`"uh huh"`, `"yeah"`, `"right"`) are classified as backchannels and suppressed.
+3. On silence without terminal punctuation, after `minSilenceBeforeCheckMs` (default 500 ms), an LLM is queried: "Is this utterance a complete thought?" Results are LRU-cached.
+   - `COMPLETE` → turn ends, reason `semantic_model`.
+   - `INCOMPLETE` → waiting continues; eventual silence timeout acts as final fallback.
+   - `TIMEOUT` → falls back to silence timeout.
+
+## Configuration
+
+```json
+{
+  "voice": {
+    "endpointing": "semantic",
+    "endpointingOptions": {
+      "model": "gpt-4o-mini",
+      "timeoutMs": 500,
+      "minSilenceBeforeCheckMs": 500,
+      "silenceTimeoutMs": 2000
+    }
+  }
+}
+```
+
+## Provider Rules
+
+- Use `gpt-4o-mini` (or equivalent cheap small model) for the classifier — latency matters more than quality for this binary decision.
+- Keep `timeoutMs` under 600 ms to avoid adding noticeable lag to the turn boundary.
+- Increase `silenceTimeoutMs` for users who speak slowly or pause frequently.
+- Reduce `minSilenceBeforeCheckMs` to 300 ms for faster-paced conversations.
+
+## Events
+
+| Event                  | Description                                                              |
+|------------------------|--------------------------------------------------------------------------|
+| `turn_complete`        | User turn ended; `reason` is `punctuation`, `semantic_model`, or `silence_timeout` |
+| `backchannel_detected` | A backchannel phrase was recognised; accumulation suppressed             |
+
+## Examples
+
+- "Use semantic endpoint detection to avoid cutting me off mid-thought."
+- "Enable smarter turn detection for this conversational voice session."
+- "Configure the endpoint detector to wait longer before deciding I'm done speaking."
+
+## Constraints
+
+- Requires an LLM provider to be configured in the runtime for the classifier calls.
+- LLM calls add latency at turn boundaries. Use a small, fast model to minimize this.
+- The LRU cache (keyed on first 100 characters) reduces repeated LLM calls for identical short utterances.

package/registry/curated/google-cloud-stt/SKILL.md

@@ -0,0 +1,71 @@
+---
+name: google-cloud-stt
+version: '1.0.0'
+description: Batch speech-to-text via Google Cloud Speech-to-Text API — LINEAR16 PCM, configurable language, word-level confidence scores.
+author: Wunderland
+namespace: wunderland
+category: voice
+tags: [voice, stt, speech-to-text, google, cloud, batch]
+requires_secrets: [google.cloudSttCredentials]
+requires_tools: []
+metadata:
+  agentos:
+    emoji: "\U0001F3A4"
+    primaryEnv: GOOGLE_CLOUD_STT_CREDENTIALS
+    homepage: https://cloud.google.com/speech-to-text/docs
+---
+
+# Google Cloud STT
+
+Use this skill when Google Cloud Speech-to-Text is the preferred provider, particularly when the user's Google Cloud project is already configured or when multi-language support is required without switching providers.
+
+Prefer this over Deepgram when the agent is already operating within a Google Cloud environment, or when the user specifically requests Google as the STT backend.
+
+## Setup
+
+Set `GOOGLE_CLOUD_STT_CREDENTIALS` in the environment or agent secrets store. Accepts either:
+- An absolute path to a service-account JSON key file
+- A raw JSON string with the service-account credentials
+
+## Configuration
+
+```json
+{
+  "voice": {
+    "stt": "google-cloud-stt"
+  }
+}
+```
+
+With language override:
+
+```json
+{
+  "voice": {
+    "stt": "google-cloud-stt",
+    "providerOptions": {
+      "language": "fr-FR"
+    }
+  }
+}
+```
+
+## Provider Rules
+
+- Language codes follow BCP-47 format (e.g. `en-US`, `fr-FR`, `ja-JP`). Default is `en-US`.
+- Audio input must be LINEAR16 PCM format. Convert other formats before sending.
+- Confidence scores and word-level alternatives are included in the transcription result.
+- Credentials are resolved at initialization; if the path is invalid or JSON is malformed, initialization will throw immediately.
+
+## Examples
+
+- "Transcribe this audio using Google Cloud Speech-to-Text."
+- "Use Google STT with French language recognition."
+- "Set up Google Cloud STT with my service account credentials."
+
+## Constraints
+
+- Requires a Google Cloud service account with the `Speech-to-Text` API enabled.
+- Credentials must be either a path to a valid JSON key file or inline JSON string.
+- This is a batch (non-streaming) provider. For real-time streaming, use Deepgram or Whisper instead.
+- API costs apply per 15-second audio block.

package/registry/curated/google-cloud-tts/SKILL.md

@@ -0,0 +1,71 @@
+---
+name: google-cloud-tts
+version: '1.0.0'
+description: Text-to-speech synthesis via Google Cloud Text-to-Speech API — MP3 output, configurable language and voice, voice listing.
+author: Wunderland
+namespace: wunderland
+category: voice
+tags: [voice, tts, text-to-speech, google, cloud, neural]
+requires_secrets: [google.cloudTtsCredentials]
+requires_tools: []
+metadata:
+  agentos:
+    emoji: "\U0001F50A"
+    primaryEnv: GOOGLE_CLOUD_TTS_CREDENTIALS
+    homepage: https://cloud.google.com/text-to-speech/docs
+---
+
+# Google Cloud TTS
+
+Use this skill when Google Cloud Text-to-Speech is the preferred synthesis provider, particularly within a Google Cloud environment or when the user needs one of Google's Neural2 or WaveNet voices not available through other providers.
+
+Prefer this over OpenAI TTS when the user already uses Google Cloud, needs a specific non-English language voice, or prefers Google's voice catalog.
+
+## Setup
+
+Set `GOOGLE_CLOUD_TTS_CREDENTIALS` in the environment or agent secrets store. Accepts either:
+- An absolute path to a service-account JSON key file
+- A raw JSON string with the service-account credentials
+
+## Configuration
+
+```json
+{
+  "voice": {
+    "tts": "google-cloud-tts"
+  }
+}
+```
+
+With voice and language options:
+
+```json
+{
+  "voice": {
+    "tts": "google-cloud-tts",
+    "providerOptions": {
+      "languageCode": "en-GB",
+      "voice": "en-GB-Neural2-A"
+    }
+  }
+}
+```
+
+## Provider Rules
+
+- Output is MP3 (audio/mpeg). Playback requires an MP3-capable audio pipeline.
+- Language codes follow BCP-47 (e.g. `en-US`, `en-GB`, `de-DE`).
+- Call `listAvailableVoices()` to enumerate all voices available on the account; useful for letting the user pick a voice.
+- Neural2 and WaveNet voices require the respective API feature to be enabled on the Google Cloud project.
+
+## Examples
+
+- "Use Google Cloud TTS with a British English Neural2 voice."
+- "List the available voices on my Google Cloud TTS account."
+- "Synthesize this response using Google Cloud Text-to-Speech."
+
+## Constraints
+
+- Requires a Google Cloud service account with the `Text-to-Speech` API enabled.
+- Credentials must be either a path to a valid JSON key file or inline JSON string.
+- API costs apply per character synthesized.

package/registry/curated/image-editing/SKILL.md

@@ -0,0 +1,25 @@
+---
+name: image-editing
+description: Edit, transform, and enhance images using AI models
+version: 1.0.0
+tags: [image, editing, img2img, inpainting, upscaling, variation]
+tools_required: [editImage, upscaleImage, variateImage]
+---
+
+# Image Editing
+
+Edit images with img2img transformation, inpainting (fill masked regions), outpainting (extend beyond borders), upscaling (2x/4x super resolution), and creating variations.
+
+## Capabilities
+- **img2img**: Transform an image based on a text prompt
+- **Inpainting**: Fill in masked regions of an image
+- **Upscaling**: Increase image resolution by 2x or 4x
+- **Variations**: Create variations of an existing image
+
+## Providers
+OpenAI, Stability AI, Local Stable Diffusion (A1111/ComfyUI), Replicate
+
+## Example
+"Take this photo and make it look like a watercolor painting"
+"Remove the background from this product photo"
+"Upscale this image to 4x resolution"