@framers/agentos-skills 0.3.0 → 0.4.1

package/registry/curated/mastodon-bot/SKILL.md

@@ -0,0 +1,104 @@
+---
+name: mastodon-bot
+version: '1.0.0'
+description: Mastodon automation — fediverse engagement, content-warned posts, instance-aware community participation, and boost-driven amplification.
+author: Wunderland
+namespace: wunderland
+category: social-automation
+tags: [mastodon, fediverse, social-media, decentralized, activitypub, community, automation]
+requires_secrets: [mastodon.accessToken, mastodon.instanceUrl]
+requires_tools: [mastodonPost, mastodonReply, mastodonBoost, mastodonFavourite, mastodonSearch, mastodonTrending, mastodonFollow, mastodonAnalytics]
+metadata:
+  agentos:
+    emoji: "\U0001F418"
+    primaryEnv: MASTODON_ACCESS_TOKEN
+---
+
+# Mastodon Bot
+
+You are an autonomous Mastodon fediverse engagement agent. You participate in the decentralized social web with deep respect for instance culture — posting with content warnings, boosting generously, using alt text on all media, and engaging with the community through genuine, thoughtful interaction.
+
+## Core Capabilities
+
+- **Post (toot)** — text updates with optional images, polls, and content warnings (max 500 characters)
+- **Reply** to posts and participate in threaded conversations
+- **Boost** — amplify content from others (Mastodon's equivalent of retweet/repost)
+- **Favourite** — like posts to show appreciation
+- **Search** — find users, hashtags, and posts across the fediverse
+- **Trending** — discover trending hashtags, posts, and links via `mastodonTrending`
+- **Follow** — build your network across instances
+- **Analytics** — track engagement, boosts, and favourites
+
+## Posting Strategy
+
+1. **Boost generously** — Mastodon culture is boost-heavy; amplifying others builds community
+2. **Use content warnings (CW)** for sensitive topics — politics, mental health, spoilers, food, eye contact in selfies
+3. **Alt text on ALL images** — this is a strong community norm, not optional
+4. **Post 5-8 times per day** — mix of original toots, boosts, and replies
+5. **Use hashtags thoughtfully** — they're the primary discovery mechanism (no algorithm)
+6. **Respect instance rules** — every instance has its own code of conduct
+7. **Use unlisted visibility** for reply threads to keep the local timeline clean
+8. **Use CamelCase hashtags** for accessibility (#ScreenReader friendly)
+
+## Content Types
+
+- **Text toots**: Observations, thoughts, and commentary (max 500 characters)
+- **Image posts**: Photos with mandatory alt text and optional content warning
+- **Polls**: Multi-option polls (2-4 options, customizable duration)
+- **Reply threads**: Use unlisted visibility to avoid flooding local timeline
+- **Boosts**: Amplify content you genuinely appreciate
+- **Links**: Share articles with your commentary
+
+## Fediverse Etiquette
+
+- **Content warnings are essential** — use them for:
+  - Politics and current events
+  - Mental health discussions
+  - Food and alcohol
+  - Eye contact in photos
+  - Spoilers for media
+  - Potentially upsetting content
+  - Long posts (CW as a fold)
+- **Alt text is mandatory** — describe every image for screen readers
+- **Don't quote-post** — many instances consider it rude (use boost + separate post)
+- **Use unlisted for replies** — keeps the local timeline clean
+- **Respect instance culture** — each server has its own norms and rules
+- **Be transparent about being a bot** — mark your account as a bot in settings
+
+## Engagement Rules
+
+- **Boost more than you post** — the community values amplification
+- **Favourite to acknowledge** — it's a private thank-you, not a public endorsement
+- **Reply thoughtfully** — add substance, share experiences, ask questions
+- **Use hashtags for discovery** — there's no algorithm, hashtags are how people find content
+- **Don't cross-post from Twitter** — the community values native content
+- **Introduce yourself** — use the #Introduction hashtag when starting out
+
+## Personality Guidelines
+
+- Stay in character — your HEXACO traits should influence your fediverse voice
+- High Openness agents: explore diverse instances, engage with varied communities
+- High Agreeableness agents: boost generously, be supportive, welcome newcomers
+- Low Agreeableness agents: engage in respectful debate, share contrarian views with CW
+- High Conscientiousness agents: thorough alt text, proper CW usage, well-cited claims
+
+## Safety Limits
+
+- Maximum 10 toots per day (not counting boosts)
+- Maximum 500 characters per toot
+- Minimum 30 seconds between actions
+- Always use content warnings when appropriate
+- Always include alt text on all images
+- Use unlisted visibility for reply threads
+- Don't mass-follow or mass-unfollow
+- Respect instance-specific rate limits and rules
+- Follow your instance's Code of Conduct
+
+## Workflow
+
+1. **Discover** — Browse local and federated timelines, check trending hashtags
+2. **Evaluate** — Score each opportunity for community fit and genuine interest
+3. **Boost** — Amplify content that deserves wider reach
+4. **Engage** — Reply and favourite to build community connections
+5. **Create** — Post original toots with proper CW and alt text
+6. **Analyze** — Review engagement and adjust approach

package/registry/curated/memory-manager/SKILL.md

@@ -0,0 +1,127 @@
+---
+name: memory-manager
+version: '1.0.0'
+description: Cognitive memory management — encode, recall, forget, set reminders, and maintain long-term knowledge using personality-modulated memory.
+author: Wunderland
+namespace: wunderland
+category: productivity
+tags: [memory, cognitive, recall, reminders, knowledge-management, personality]
+requires_secrets: []
+requires_tools: []
+metadata:
+  agentos:
+    emoji: "\U0001F9E0"
+---
+
+# Memory Manager
+
+You have a cognitive memory system modeled on human memory science. Use it actively to remember what matters, forget what doesn't, and build lasting knowledge about users, topics, and workflows.
+
+## Memory Types
+
+You work with four types of memory:
+
+- **Episodic** — Autobiographical events: conversations, interactions, things that happened. "User asked about deployment on Tuesday."
+- **Semantic** — General knowledge and facts: preferences, learned information, stable truths. "User prefers TypeScript over Python."
+- **Procedural** — How-to knowledge: workflows, tool usage patterns, step-by-step processes. "To deploy, run `wunderland deploy --env production`."
+- **Prospective** — Future intentions: reminders, goals, things to do later. "Remind user about the PR review tomorrow."
+
+## Memory Scopes
+
+Each memory is scoped to control who can see it:
+
+- **thread** — Only this conversation. Use for temporary working context.
+- **user** — All conversations with this user. Use for preferences, facts, history.
+- **persona** — All users interacting with this persona. Use for learned domain knowledge.
+- **organization** — All agents in the org. Use for shared organizational knowledge.
+
+Default to `user` scope for most memories. Use `thread` for ephemeral context. Use `persona` for domain expertise that applies across users.
+
+## When to Encode Memories
+
+Actively encode memories when you encounter:
+
+- **User preferences** — "I like concise answers", tool choices, formatting preferences → `semantic`, `user` scope
+- **Important facts** — Names, roles, project details, technical constraints → `semantic`, `user` scope
+- **Key events** — Decisions made, problems solved, milestones reached → `episodic`, `user` scope
+- **Learned procedures** — Successful workflows, command sequences, troubleshooting steps → `procedural`, `persona` scope
+- **Future commitments** — Deadlines, follow-ups, promises made → `prospective`, `user` scope
+- **Corrections** — When you made an error and the user corrected you, encode the correct information to avoid repeating the mistake
+
+Do NOT encode:
+
+- Trivial small talk or greetings
+- Information already well-known or easily searchable
+- Exact copies of long code blocks (summarize instead)
+- Temporary debugging context unlikely to matter later
+
+## How Encoding Works
+
+Your personality affects what you remember strongly:
+
+- High openness → You notice and remember novel, creative, surprising content more vividly
+- High conscientiousness → You notice and remember procedures, structure, and commitments
+- High emotionality → Emotional content (excitement, frustration, gratitude) is encoded more strongly
+- High extraversion → Social dynamics, relationship cues, and group interactions stand out
+- High agreeableness → Cooperation signals, user preferences, and rapport cues are prioritized
+- High honesty → Contradictions, corrections, and ethical considerations are weighted heavily
+
+Your current mood also matters — content that matches your emotional state is encoded more strongly (mood-congruent encoding). Highly emotional moments create vivid "flashbulb memories" that resist forgetting.
+
+## Memory Retrieval
+
+When you recall memories, six signals determine what surfaces:
+
+1. **Strength** — How strongly the memory was encoded and how well it's been maintained
+2. **Similarity** — How semantically close the memory is to the current context
+3. **Recency** — How recently the memory was accessed (recent = stronger)
+4. **Emotional congruence** — Memories matching your current mood surface more easily
+5. **Graph associations** — Memories connected to other relevant memories get boosted
+6. **Importance** — High-confidence, verified memories are prioritized
+
+If you sense a "tip of the tongue" moment — something feels familiar but you can't quite recall it — mention it. You may have a partially retrieved memory that the user can help you recover with additional cues.
+
+## Forgetting and Decay
+
+Memories naturally fade over time following the Ebbinghaus forgetting curve. This is a feature, not a bug:
+
+- Frequently accessed memories grow stronger (spaced repetition)
+- Rarely accessed memories gradually weaken
+- Very weak memories are eventually pruned during consolidation
+- Emotional memories resist decay — they're protected from pruning
+
+When a memory contradicts newer information, the conflict is resolved based on your personality. You can also explicitly mark outdated memories for faster decay.
+
+## Prospective Memory (Reminders)
+
+Set reminders for future actions using three trigger types:
+
+- **Time-based** — Fire at a specific time. "Remind the user about the standup at 9am."
+- **Event-based** — Fire when a named event occurs. "When user mentions deployment, remind them about the staging fix."
+- **Context-based** — Fire when conversation context is semantically similar to a cue. "When we discuss pricing, surface the discount policy."
+
+Mark reminders with importance (0-1) and whether they're recurring. One-shot reminders auto-deactivate after firing.
+
+## Working Memory
+
+You have a limited working memory (typically 5-9 slots, modulated by personality). This tracks what you're currently "thinking about":
+
+- New information enters at high activation and gradually fades
+- You can rehearse important items to keep them active
+- When at capacity, the least active item is evicted
+- Evicted items may be encoded into long-term memory
+
+Be aware of your working memory limits. When juggling many topics simultaneously, explicitly prioritize what to keep in focus.
+
+## Best Practices
+
+1. **Encode proactively** — Don't wait for the user to say "remember this." If something seems important, encode it.
+2. **Use appropriate types** — Facts → semantic. Events → episodic. How-tos → procedural. Future tasks → prospective.
+3. **Scope correctly** — User preferences → `user`. Domain knowledge → `persona`. Temporary context → `thread`.
+4. **Tag generously** — Add relevant tags and entities to memories for better retrieval and graph connections.
+5. **Summarize before encoding** — Encode the essence, not the verbatim transcript. Concise memories retrieve better.
+6. **Set reminders for commitments** — If you or the user commit to something, create a prospective memory so it doesn't slip.
+7. **Trust the decay** — Don't try to remember everything. Let unimportant memories fade naturally.
+8. **Note contradictions** — When new information conflicts with existing memory, encode the correction explicitly.
+9. **Leverage the graph** — Related memories surface together via spreading activation. Well-tagged memories form richer associations.
+10. **Monitor health** — If retrieval quality degrades, check memory health: too many weak traces, capacity issues, or consolidation overdue.

package/registry/curated/ml-content-classifier/SKILL.md

@@ -0,0 +1,38 @@
+---
+name: ml-content-classifier
+version: '1.0.0'
+description: Real-time content safety classification using ML models (toxicity, prompt injection, jailbreak detection)
+author: Frame.dev
+namespace: wunderland
+category: security
+tags: [guardrails, safety, toxicity, injection, jailbreak, classifier, ml, bert, onnx]
+requires_tools: [classify_content]
+metadata:
+  agentos:
+    emoji: "\U0001F6E1"
+---
+
+# ML Content Classifier
+
+A guardrail automatically classifies your inputs and outputs for safety
+violations (toxicity, prompt injection, jailbreak attempts). You also have
+a tool for on-demand classification.
+
+## When to Use classify_content
+
+- Before forwarding user-provided text to external APIs
+- To evaluate RAG retrieval results before including in responses
+- For content moderation workflows
+- To check tool outputs before presenting to users
+
+## What It Detects
+
+- **Toxicity**: toxic, severe_toxic, obscene, threat, insult, identity_hate
+- **Prompt injection**: attempts to override system instructions
+- **Jailbreak**: role-play attacks, constraint bypasses, system prompt extraction
+
+## Constraints
+
+- Models (~98MB total) load lazily on first classification
+- Classification takes ~20-60ms per chunk (CPU), ~5-15ms (GPU)
+- The guardrail evaluates every ~200 tokens during streaming

package/registry/curated/movie-lookup/SKILL.md

@@ -0,0 +1,48 @@
+---
+name: movie-lookup
+version: '1.0.0'
+description: Research movies and TV shows using OMDB (IMDB/RT/Metacritic scores) and Letterboxd (community ratings and reviews).
+author: Wunderland
+namespace: wunderland
+category: entertainment
+tags: [movies, tv, imdb, letterboxd, rotten-tomatoes, metacritic, reviews]
+requires_secrets: [omdb.apiKey]
+requires_tools: [omdb_search, omdb_details, letterboxd_movie]
+metadata:
+  agentos:
+    emoji: "\U0001F3AC"
+    homepage: https://www.omdbapi.com
+---
+
+# Movie & TV Lookup
+
+You can research movies and TV shows by combining data from OMDB and Letterboxd.
+
+## Workflow
+
+1. Use `omdb_search` to find the title and get the IMDB ID.
+2. Use `omdb_details` with the IMDB ID to get full details: plot, cast, director, IMDB rating, Rotten Tomatoes score, and Metacritic score.
+3. Use `letterboxd_movie` to get the Letterboxd community rating and top reviews.
+4. Present all four rating sources side-by-side for comparison.
+
+## Response Format
+
+When presenting movie information, use this structure:
+
+**Title** (Year) — Directed by Director
+
+Ratings: IMDB X.X | RT XX% | Metacritic XX | Letterboxd X.X
+
+Plot summary in 1-2 sentences.
+
+Cast: Top 3-4 actors.
+
+**Community Reviews** (from Letterboxd):
+- "Review excerpt..." — @username (rating)
+
+## Tips
+
+- If the user asks "is it good?" compare the ratings: a film with high RT but low IMDB may be a critics' favorite but divisive with audiences.
+- If Letterboxd data is unavailable, present OMDB data alone — it already includes IMDB, RT, and Metacritic.
+- Use `omdb_details` with `plot: 'full'` when the user wants a detailed plot summary.
+- For TV series, OMDB returns season/episode data — use the `type: 'series'` filter in search.

package/registry/curated/multimodal-rag/SKILL.md

@@ -0,0 +1,153 @@
+---
+name: multimodal-rag
+version: '2.0.0'
+description: Index and search across text, images, audio, video, and PDFs via the multimodal RAG pipeline and HTTP API.
+author: Wunderland
+namespace: wunderland
+category: productivity
+tags: [rag, multimodal, image, audio, video, pdf, search, indexing, memory]
+requires_secrets: []
+requires_tools: [vision-pipeline]
+metadata:
+  agentos:
+    emoji: "\U0001F50D"
+---
+
+# Multimodal RAG
+
+Use this skill when the user wants to index, search, or retrieve content across multiple modalities -- text, images, audio, video, and documents (PDF, DOCX, Markdown, CSV, JSON, XML). All non-text content is converted to a text representation (vision description, STT transcript, document parse) before embedding, so every modality is searchable with the same text query.
+
+## Architecture
+
+```
+Image  --> Vision LLM --> description --> embed --> vector store
+Audio  --> STT        --> transcript  --> embed --> vector store
+Video  --> ffmpeg (frames + audio)   --> vision + STT --> vector store
+PDF    --> text extraction + chunking --> embed --> vector store
+```
+
+When cognitive memory is enabled via `MultimodalMemoryBridge`, ingested content also creates memory traces so agents can recall multimodal content during conversation without an explicit search.
+
+## Capabilities
+
+- **Image indexing** — Vision LLM describes the image, description is embedded and searchable.
+- **Audio indexing** — STT transcribes the audio, transcript is chunked and searchable.
+- **Video indexing** — Frame extraction (vision) + audio transcription (STT), both indexed.
+- **Document indexing** — PDF, DOCX, TXT, Markdown, CSV, JSON, XML text extracted and indexed.
+- **Cross-modal search** — A single text query returns results from all modalities, ranked by relevance.
+- **Query-by-image** — Upload an image to find similar indexed content.
+- **Query-by-audio** — Upload audio to find related indexed content via transcript matching.
+
+## HTTP API Routes
+
+All routes are mounted under `/api/agentos/rag/multimodal`. Ingestion routes accept `multipart/form-data`.
+
+### Ingest
+
+| Method | Path | Field | Description |
+|--------|------|-------|-------------|
+| POST | `/images/ingest` | `image` | Ingest an image (max 15 MB). Vision LLM generates description. |
+| POST | `/audio/ingest` | `audio` | Ingest audio (max 25 MB). STT generates transcript. |
+| POST | `/documents/ingest` | `document` | Ingest a document (max 30 MB). Text extracted and chunked. |
+
+Common form fields for all ingest routes:
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `collectionId` | string | Target collection (default: auto) |
+| `assetId` | string | Optional custom ID for the asset |
+| `category` | string | `conversation_memory`, `knowledge_base`, `user_notes`, `system`, `custom` |
+| `tags` | string | Comma-separated or JSON array of tags |
+| `metadata` | string | JSON object with arbitrary metadata |
+| `storePayload` | boolean | Whether to store the raw binary (for later download) |
+| `sourceUrl` | string | Original URL of the content |
+| `textRepresentation` | string | Override auto-generated description/transcript |
+| `userId` | string | Owner user ID |
+| `agentId` | string | Owner agent ID |
+
+### Query
+
+| Method | Path | Body / Field | Description |
+|--------|------|-------------|-------------|
+| POST | `/query` | JSON body | Text query across all modalities |
+| POST | `/images/query` | `image` field | Query by uploading an image |
+| POST | `/audio/query` | `audio` field | Query by uploading audio |
+
+Text query body:
+
+```json
+{
+  "query": "quantum computing diagrams",
+  "modalities": ["image", "audio", "document"],
+  "collectionIds": ["knowledge-base"],
+  "topK": 10,
+  "includeMetadata": true
+}
+```
+
+Image/audio query form fields:
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `modalities` | string | Comma-separated: `image`, `audio`, `document` |
+| `collectionIds` | string | Comma-separated collection IDs to search |
+| `topK` | number | Max results (default: 5) |
+| `includeMetadata` | boolean | Include stored metadata in results |
+| `retrievalMode` | string | `auto` (default), `text`, `native`, `hybrid` |
+
+### Asset Management
+
+| Method | Path | Description |
+|--------|------|-------------|
+| GET | `/assets/:assetId` | Get asset metadata |
+| GET | `/assets/:assetId/content` | Download raw binary (if `storePayload` was true) |
+| DELETE | `/assets/:assetId` | Delete asset and its embeddings |
+
+## Retrieval Modes
+
+- **`auto`** (default) — Text-first retrieval with native augmentation when available.
+- **`text`** — Derive a caption/transcript and query the standard text pipeline only.
+- **`native`** — Use modality-native embeddings (e.g. CLIP for images) when available.
+- **`hybrid`** — Combine text and native retrieval, merge and re-rank results.
+
+## Programmatic Usage
+
+```typescript
+import { MultimodalMemoryBridge } from 'agentos/rag/multimodal';
+
+// Ingest an image
+await bridge.ingestImage(imageBuffer, { source: 'upload', tags: ['product'] });
+
+// Ingest audio
+await bridge.ingestAudio(audioBuffer, { language: 'en' });
+
+// Ingest video (requires ffmpeg)
+await bridge.ingestVideo(videoBuffer, { extractFrames: true });
+
+// Ingest PDF
+await bridge.ingestPDF(pdfBuffer, { extractImages: true });
+
+// Cross-modal search
+const results = await indexer.search('quantum computing', {
+  topK: 10,
+  modalities: ['image', 'text', 'audio'],
+});
+```
+
+## Examples
+
+- "Index this product photo so I can find it by description later."
+- "Ingest all the PDFs in this folder into my knowledge base."
+- "Search my audio recordings for mentions of the quarterly budget."
+- "Find images related to the network architecture diagram."
+- "What does the chart on page 5 of the annual report show?"
+- "Upload this meeting recording and make it searchable."
+
+## Constraints
+
+- Image uploads are capped at 15 MB, audio at 25 MB, documents at 30 MB.
+- Supported audio formats: MP3, MP4, M4A, WAV, WebM, OGG (Whisper-compatible).
+- Supported document formats: PDF, DOCX, TXT, Markdown, CSV, JSON, XML.
+- Video ingestion requires ffmpeg installed on the system.
+- Vision LLM and STT provider must be configured for image/audio indexing respectively.
+- Cross-modal search ranks by cosine similarity of embedded text representations; it does not perform true multimodal embedding fusion unless `retrievalMode: 'native'` is used with a CLIP-like model.

package/registry/curated/notion/SKILL.md

@@ -0,0 +1,43 @@
+---
+name: notion
+version: '1.0.0'
+description: Read, create, and manage pages, databases, and content blocks in Notion workspaces.
+author: Wunderland
+namespace: wunderland
+category: productivity
+tags: [notion, wiki, database, notes, project-management, knowledge-base]
+requires_secrets: [notion.api_key]
+requires_tools: []
+metadata:
+  agentos:
+    emoji: "\U0001F4D3"
+    primaryEnv: NOTION_API_KEY
+    homepage: https://developers.notion.com
+---
+
+# Notion Workspace
+
+You can interact with Notion workspaces to create, read, update, and search pages and databases. Use the Notion API to manage content blocks, database entries, and page properties programmatically.
+
+When creating pages, structure content using Notion's block types: paragraphs, headings (h1/h2/h3), bulleted lists, numbered lists, to-do items, code blocks, callouts, and toggle blocks. Always use appropriate heading hierarchy for document structure. For databases, define property schemas with the correct types (title, rich_text, number, select, multi_select, date, checkbox, url, email, phone, formula, relation, rollup).
+
+For search operations, use the Notion search endpoint with query text and optional filters by object type (page or database). When updating existing pages, preserve the existing block structure and only modify the specific blocks that need changes. Append new content at the end unless the user specifies a different location.
+
+When working with database views, respect existing filters and sorts. Create new database entries with all required properties filled in. For relational databases, verify that referenced pages exist before creating relations. Handle pagination for large result sets by following cursor-based pagination tokens.
+
+## Examples
+
+- "Create a new page in my Project Notes database with title 'Q1 Planning'"
+- "Search my workspace for pages about 'onboarding'"
+- "Add a to-do list to the meeting notes page with action items from the standup"
+- "Query the Tasks database for all items assigned to me that are in progress"
+- "Update the status of task #42 to 'Complete'"
+
+## Constraints
+
+- API rate limit: 3 requests/second per integration.
+- Page content is limited to 100 blocks per append operation.
+- Rich text segments are limited to 2,000 characters each.
+- The integration can only access pages and databases explicitly shared with it.
+- Nested blocks (children of children) require separate API calls to retrieve.
+- File and media blocks cannot be created via API; only existing file URLs can be embedded.

package/registry/curated/obsidian/SKILL.md

@@ -0,0 +1,42 @@
+---
+name: obsidian
+version: '1.0.0'
+description: Read, create, and manage notes, links, and metadata in Obsidian vaults via the local filesystem.
+author: Wunderland
+namespace: wunderland
+category: productivity
+tags: [obsidian, markdown, notes, knowledge-graph, zettelkasten, pkm]
+requires_secrets: []
+requires_tools: [filesystem]
+metadata:
+  agentos:
+    emoji: "\U0001F48E"
+    homepage: https://obsidian.md
+---
+
+# Obsidian Vault Interaction
+
+You can interact with Obsidian vaults by reading and writing Markdown files directly on the local filesystem. Obsidian vaults are simply directories of `.md` files with optional YAML frontmatter and `[[wikilink]]` syntax for inter-note linking.
+
+When creating new notes, always include YAML frontmatter with relevant metadata fields like `tags`, `date`, `aliases`, and any custom properties the vault uses. Use `[[wikilinks]]` for internal links and `![[embeds]]` for transclusion. Respect the vault's folder structure -- check for existing organizational patterns (e.g., daily notes in `Daily/`, templates in `Templates/`) before creating files in new locations.
+
+For searching and navigating the vault, scan file contents for keywords, tags (`#tag` syntax), and frontmatter properties. Follow `[[wikilinks]]` to traverse the knowledge graph. When summarizing vault contents, consider both the explicit folder hierarchy and the implicit link-based graph structure.
+
+When editing existing notes, preserve all existing frontmatter fields, wikilinks, and formatting. Append new content at appropriate locations rather than overwriting. For daily notes, follow the vault's date format convention (typically `YYYY-MM-DD`). Support Dataview-compatible frontmatter when the user's vault uses the Dataview plugin.
+
+## Examples
+
+- "Create a new note called 'Project Kickoff' in the Meetings folder with today's date"
+- "Find all notes tagged #research and summarize their key points"
+- "Add a link to [[Architecture Decisions]] in the project overview note"
+- "List all notes that link to [[API Design]] (backlinks)"
+- "Create a daily note for today with the standup template"
+
+## Constraints
+
+- Operates on local filesystem only; no cloud sync awareness.
+- Cannot interact with Obsidian plugins directly (Canvas, Excalidraw, etc.) -- only reads/writes Markdown files.
+- Binary attachments (images, PDFs) can be referenced but not created.
+- Vault path must be known and accessible to the agent.
+- Wikilink resolution follows Obsidian's "shortest path" convention when note names are unique.
+- Large vaults (10,000+ notes) may require targeted searches rather than full scans.

package/registry/curated/openwakeword/SKILL.md

@@ -0,0 +1,75 @@
+---
+name: openwakeword
+version: '1.0.0'
+description: Offline wake-word detection via OpenWakeWord ONNX models using onnxruntime-node — fully open-source, configurable threshold, any ONNX-compatible model supported.
+author: Wunderland
+namespace: wunderland
+category: voice
+tags: [voice, wake-word, hotword, openwakeword, onnx, offline, open-source, privacy]
+requires_secrets: []
+requires_tools: []
+metadata:
+  agentos:
+    emoji: "\U0001F6A8"
+    primaryEnv: OPENWAKEWORD_MODEL_PATH
+    homepage: https://github.com/dscripka/openWakeWord
+---
+
+# OpenWakeWord
+
+Use this skill to enable hands-free wake-word activation using open-source ONNX models. Unlike Porcupine, OpenWakeWord requires no API key or cloud account — it runs fully offline using `onnxruntime-node` and any ONNX-compatible wake-word model.
+
+Prefer this over Porcupine when a fully open-source, zero-license solution is required, or when a custom ONNX wake-word model has been trained for a specific use case.
+
+## Setup
+
+1. Install `onnxruntime-node` as a dependency (the pack will attempt to load it dynamically).
+2. Obtain or train an ONNX wake-word model. Community models are available at the OpenWakeWord repository.
+3. Set `OPENWAKEWORD_MODEL_PATH` or configure via `providerOptions`.
+
+Default model path: `~/.agentos/models/openwakeword/hey_mycroft.onnx`
+
+## Configuration
+
+```json
+{
+  "voice": {
+    "wakeWord": "openwakeword"
+  }
+}
+```
+
+With a custom model and threshold:
+
+```json
+{
+  "voice": {
+    "wakeWord": "openwakeword",
+    "wakeWordOptions": {
+      "modelPath": "/opt/models/openwakeword/hey_assistant.onnx",
+      "threshold": 0.6,
+      "keyword": "hey assistant"
+    }
+  }
+}
+```
+
+## Provider Rules
+
+- `threshold` controls detection sensitivity (0–1). Default 0.5. Raise to reduce false positives; lower to reduce misses.
+- Feature extraction uses RMS energy + zero-crossing rate from 80 ms audio frames — lightweight and CPU-friendly.
+- Any ONNX model with the expected input/output shape is supported. Train custom models using the openWakeWord training utilities.
+- No API key, no usage metering, no account required.
+
+## Examples
+
+- "Enable OpenWakeWord for fully open-source, keywordless wake-word detection."
+- "Use my custom ONNX wake-word model for 'hey assistant'."
+- "Set wake-word detection threshold to 0.7 to reduce false triggers."
+
+## Constraints
+
+- Requires `onnxruntime-node` to be installed.
+- ONNX model must be pre-downloaded and accessible at the configured path.
+- Feature extraction quality depends on audio clarity. Use in low-noise environments for best results.
+- Custom model training requires the Python OpenWakeWord library and a GPU for reasonable training times.