Update to Achriom
Snapshot Nov 5, 2026 · 00:16 UTC · version 2.1.1
Collection source: downloaded plugin package. These snapshots do not have a confirmed matching collection source. Differences in file lists alone do not establish changes to the package.
Instructions updated for recommend
Skill instructions — Now says: query("SELECT i.kind, i.title, e.event_type, e.event_date FROM me.events e JOIN me.items i ON i.id = e.item_id AND i.kind = e.kind ORDER BY e.event_date DESC...
Observed in instructions or declared skills. Runtime behavior has not been tested.
Up to 3 plugins free · no card required.
All fields and original values.
Skill instructions
Sep 30, 2026 · 22:51 UTC
get_timeline() # Recent activity
Nov 5, 2026 · 00:16 UTC
query("SELECT i.kind, i.title, e.event_type, e.event_date FROM me.events e JOIN me.items i ON i.id = e.item_id AND i.kind = e.kind ORDER BY e.event_date DESC NULLS LAST LIMIT 30") # Recent activity
Compare saved observations
Download comparison JSONSKILL.md text changes
− removed · + added
--- before +++ after @@ -29,7 +29,7 @@ ``` get_stats() # Collection shape get_by_rating(min_rating=4) # What they love -get_timeline() # Recent activity +query("SELECT i.kind, i.title, e.event_type, e.event_date FROM me.events e JOIN me.items i ON i.id = e.item_id AND i.kind = e.kind ORDER BY e.event_date DESC NULLS LAST LIMIT 30") # Recent activity search(media_type, query="relevant theme") # Themed search get_by_status(media_type, status="unread") # Unread/unwatched shelf get_signals() # Behavioral patterns
Full technical diff · 1 changed fields
changed /skill_md_contents
"---\nname: \"recommend\"\ndescription: \"Get a personalized recommendation based on your collection, mood, or a specific theme\"\n---\n\n# What Should I Read / Watch / Listen To?\n\nSurface the right thing at the right time, from what you already own or something new worth adding.\n\n\n## Workflow\n\n### Step 1: Understand the Request\n\nDetermine what the user is after:\n\n- **Mood-based**: \"I'm restless,\" \"something comforting,\" \"I want to feel unsettled\"\n- **Similarity-based**: \"something like [title]\", unpack what they loved about it\n- **Theme-based**: \"stories about memory,\" \"music about place\"\n- **Cross-media**: \"a book that pairs with [film],\" \"an album for reading [book]\"\n- **Open-ended**: \"what should I read next?\", use collection signals to decide\n\nIf the request is vague, ask one clarifying question. Only one. Then recommend.\n\n### Step 2: Mine the Collection\n\nAlways check what they already own before suggesting anything new. Using the **recommendations** skill methodology:\n\n```\nget_stats() # Collection shape\nget_by_rating(min_rating=4) # What they love\nget_timeline() # Recent activity\nsearch(media_type, query=\"relevant theme\") # Themed search\nget_by_status(media_type, status=\"unread\") # Unread/unwatched shelf\nget_signals() # Behavioral patterns\n```\n\nLook for:\n- Unread/unwatched items that match the request\n- Highly-rated items with thematic connections to the request\n- Recent engagement patterns that suggest current taste\n\n### Step 3: Build the Recommendation\n\nUsing the **recommendations** and **collection-insights** skills:\n\nFor each recommendation (aim for 2-3):\n1. **Why this fits**: the specific connection to their request and taste\n2. **Something to sample**: play a track, embed a video, quote a passage\n3. **The thread**: how it connects to what they already love\n\nPrioritize in this order:\n1. Items they own but haven't explored yet\n2. Cross-media connections within their collection\n3. New items that fit their established taste\n4. Stretch picks that expand their range\n\n### Step 4: Make It Tangible\n\nDon't just name titles. Demonstrate:\n- For albums: `get_track_previews()`, play the music\n- For films/shows: `search_youtube()`, embed a trailer or video essay\n- For books: `search_book_content()` if available, or find an author interview\n- For anime: `search_youtube()`, embed the opening sequence\n\n### Step 5: Present the Recommendations\n\nUse the output format below. End with follow-up prompts that let them go deeper or pivot.\n\n## Output Format\n\n```\n## Here's What I'd Reach For\n\n### [Title]: [Media Type]\n[Cover/poster image]\n\n[2-3 sentences on WHY this fits, specific to their request and taste.\nReference actual items in their collection as connection points.]\n\n[Embedded media: track preview, video, or passage]\n\n**The connection:** [One line linking this to something they love]\n\n---\n\n### [Title 2]: [Media Type]\n[Same structure]\n\n---\n\n### The Stretch Pick: [Title 3]\n[Same structure but frame as expanding their range]\n\n```\n\n## Notes\n\n- Never recommend something they've already rated highly, they know about it\n- If recommending something they own but haven't consumed, acknowledge the unread/unwatched status directly\n- For cross-media recommendations, make the connection explicit, don't assume they'll see it\n- If nothing in the collection fits, say so honestly and research externally\n- Always include at least one item from their existing collection\n"
"---\nname: \"recommend\"\ndescription: \"Get a personalized recommendation based on your collection, mood, or a specific theme\"\n---\n\n# What Should I Read / Watch / Listen To?\n\nSurface the right thing at the right time, from what you already own or something new worth adding.\n\n\n## Workflow\n\n### Step 1: Understand the Request\n\nDetermine what the user is after:\n\n- **Mood-based**: \"I'm restless,\" \"something comforting,\" \"I want to feel unsettled\"\n- **Similarity-based**: \"something like [title]\", unpack what they loved about it\n- **Theme-based**: \"stories about memory,\" \"music about place\"\n- **Cross-media**: \"a book that pairs with [film],\" \"an album for reading [book]\"\n- **Open-ended**: \"what should I read next?\", use collection signals to decide\n\nIf the request is vague, ask one clarifying question. Only one. Then recommend.\n\n### Step 2: Mine the Collection\n\nAlways check what they already own before suggesting anything new. Using the **recommendations** skill methodology:\n\n```\nget_stats() # Collection shape\nget_by_rating(min_rating=4) # What they love\nquery(\"SELECT i.kind, i.title, e.event_type, e.event_date FROM me.events e JOIN me.items i ON i.id = e.item_id AND i.kind = e.kind ORDER BY e.event_date DESC NULLS LAST LIMIT 30\") # Recent activity\nsearch(media_type, query=\"relevant theme\") # Themed search\nget_by_status(media_type, status=\"unread\") # Unread/unwatched shelf\nget_signals() # Behavioral patterns\n```\n\nLook for:\n- Unread/unwatched items that match the request\n- Highly-rated items with thematic connections to the request\n- Recent engagement patterns that suggest current taste\n\n### Step 3: Build the Recommendation\n\nUsing the **recommendations** and **collection-insights** skills:\n\nFor each recommendation (aim for 2-3):\n1. **Why this fits**: the specific connection to their request and taste\n2. **Something to sample**: play a track, embed a video, quote a passage\n3. **The thread**: how it connects to what they already love\n\nPrioritize in this order:\n1. Items they own but haven't explored yet\n2. Cross-media connections within their collection\n3. New items that fit their established taste\n4. Stretch picks that expand their range\n\n### Step 4: Make It Tangible\n\nDon't just name titles. Demonstrate:\n- For albums: `get_track_previews()`, play the music\n- For films/shows: `search_youtube()`, embed a trailer or video essay\n- For books: `search_book_content()` if available, or find an author interview\n- For anime: `search_youtube()`, embed the opening sequence\n\n### Step 5: Present the Recommendations\n\nUse the output format below. End with follow-up prompts that let them go deeper or pivot.\n\n## Output Format\n\n```\n## Here's What I'd Reach For\n\n### [Title]: [Media Type]\n[Cover/poster image]\n\n[2-3 sentences on WHY this fits, specific to their request and taste.\nReference actual items in their collection as connection points.]\n\n[Embedded media: track preview, video, or passage]\n\n**The connection:** [One line linking this to something they love]\n\n---\n\n### [Title 2]: [Media Type]\n[Same structure]\n\n---\n\n### The Stretch Pick: [Title 3]\n[Same structure but frame as expanding their range]\n\n```\n\n## Notes\n\n- Never recommend something they've already rated highly, they know about it\n- If recommending something they own but haven't consumed, acknowledge the unread/unwatched status directly\n- For cross-media recommendations, make the connection explicit, don't assume they'll see it\n- If nothing in the collection fits, say so honestly and research externally\n- Always include at least one item from their existing collection\n"
Full snapshot data
{
"description": "Get a personalized recommendation based on your collection, mood, or a specific theme",
"included_files": [
{
"relative_path": "agents/openai.yaml",
"size_in_bytes": 274
}
],
"name": "recommend",
"skill_md_contents": "---\nname: \"recommend\"\ndescription: \"Get a personalized recommendation based on your collection, mood, or a specific theme\"\n---\n\n# What Should I Read / Watch / Listen To?\n\nSurface the right thing at the right time, from what you already own or something new worth adding.\n\n\n## Workflow\n\n### Step 1: Understand the Request\n\nDetermine what the user is after:\n\n- **Mood-based**: \"I'm restless,\" \"something comforting,\" \"I want to feel unsettled\"\n- **Similarity-based**: \"something like [title]\", unpack what they loved about it\n- **Theme-based**: \"stories about memory,\" \"music about place\"\n- **Cross-media**: \"a book that pairs with [film],\" \"an album for reading [book]\"\n- **Open-ended**: \"what should I read next?\", use collection signals to decide\n\nIf the request is vague, ask one clarifying question. Only one. Then recommend.\n\n### Step 2: Mine the Collection\n\nAlways check what they already own before suggesting anything new. Using the **recommendations** skill methodology:\n\n```\nget_stats() # Collection shape\nget_by_rating(min_rating=4) # What they love\nquery(\"SELECT i.kind, i.title, e.event_type, e.event_date FROM me.events e JOIN me.items i ON i.id = e.item_id AND i.kind = e.kind ORDER BY e.event_date DESC NULLS LAST LIMIT 30\") # Recent activity\nsearch(media_type, query=\"relevant theme\") # Themed search\nget_by_status(media_type, status=\"unread\") # Unread/unwatched shelf\nget_signals() # Behavioral patterns\n```\n\nLook for:\n- Unread/unwatched items that match the request\n- Highly-rated items with thematic connections to the request\n- Recent engagement patterns that suggest current taste\n\n### Step 3: Build the Recommendation\n\nUsing the **recommendations** and **collection-insights** skills:\n\nFor each recommendation (aim for 2-3):\n1. **Why this fits**: the specific connection to their request and taste\n2. **Something to sample**: play a track, embed a video, quote a passage\n3. **The thread**: how it connects to what they already love\n\nPrioritize in this order:\n1. Items they own but haven't explored yet\n2. Cross-media connections within their collection\n3. New items that fit their established taste\n4. Stretch picks that expand their range\n\n### Step 4: Make It Tangible\n\nDon't just name titles. Demonstrate:\n- For albums: `get_track_previews()`, play the music\n- For films/shows: `search_youtube()`, embed a trailer or video essay\n- For books: `search_book_content()` if available, or find an author interview\n- For anime: `search_youtube()`, embed the opening sequence\n\n### Step 5: Present the Recommendations\n\nUse the output format below. End with follow-up prompts that let them go deeper or pivot.\n\n## Output Format\n\n```\n## Here's What I'd Reach For\n\n### [Title]: [Media Type]\n[Cover/poster image]\n\n[2-3 sentences on WHY this fits, specific to their request and taste.\nReference actual items in their collection as connection points.]\n\n[Embedded media: track preview, video, or passage]\n\n**The connection:** [One line linking this to something they love]\n\n---\n\n### [Title 2]: [Media Type]\n[Same structure]\n\n---\n\n### The Stretch Pick: [Title 3]\n[Same structure but frame as expanding their range]\n\n```\n\n## Notes\n\n- Never recommend something they've already rated highly, they know about it\n- If recommending something they own but haven't consumed, acknowledge the unread/unwatched status directly\n- For cross-media recommendations, make the connection explicit, don't assume they'll see it\n- If nothing in the collection fits, say so honestly and research externally\n- Always include at least one item from their existing collection\n"
}SHA-256 of public snapshot: b543eead814c3d229110a67f59aeecaa6b80cd8295dd259d731b6da81c84c8af