Semoss

Use BEFORE starting a non-trivial task to recall prior lessons, and IMMEDIATELY when the user corrects you, accepts a non-obvious approach, or you hit a subtle gotcha — to persist the lesson for future sessions. Covers SearchMemories, GetMemory, SaveMemoryCandidate, and MarkVerified from the Agent_Memory MCP. Skip entirely if no Agent_Memory MCP is attached to the current app.

Use when compiling, building, or publishing the React app after making changes to source files — at the end of any turn that edited client code, or when the user asks to build, rebuild, or deploy. Invoke the BuildAndPublishApp tool with the project id. Do not attempt to run node, npm, pnpm, or any JavaScript build command via Bash — the sandbox blocks node execution, and BuildAndPublishApp is the only supported build path.

Use when writing code in an app that queries a relational or graph database on the platform, running SELECTs, inserts, updates, deletes, or fetching schema/table structure. Covers the SqlQuery(), SqlQueryBase64(), and GetDatabaseTableStructure() pixel commands via @semoss/sdk's runPixel, plus listing databases with MyEngines(engineTypes=["DATABASE"]). Do not use for LLM calls (see model-engine) or vector database queries.

Implementing two-step image upload + LLM pixel call (SEMOSS pattern)

Use when writing code in an app that calls an LLM, embedding model, or other model engine, OR when listing/selecting models the user has access to. Covers the LLM() and MyEngines() pixel commands via @semoss/sdk's runPixel, including prompt/completion calls, conversational history, image inputs, and parsing model responses. Do not use for vector database queries (see semoss-vector) or guardrail engines (see semoss-guardrail).

Use when writing Python in a SEMOSS app — `py/mcp_driver.py`, helper modules, or any code that runs inside the SEMOSS Python runtime. Covers what SEMOSS injects (`ROOT`, active insight context), what to import from `ai_server` / `semoss` / `smssutil`, the `@mcp_metadata` decorator for exposing MCP tools, and how Python reaches Pixel via `Insight().run_pixel(...)`. For what Pixel commands to actually run, see the `database`, `model`, and `vector` skills. Do not use for Java reactor authoring or frontend `@semoss/sdk` calls.

Use when writing code in an app that creates, lists, renames, pins, or deletes playground rooms, reads/updates room options (model, system prompt, MCPs, temperature), or reads chat history from a room. Covers CreatePlaygroundRoom, GetPlaygroundRooms, GetWorkspaceRooms, RenameRoom, PinRoom, RemoveUserRoom, SetRoomForInsight, GetRoomOptions, UpdateRoomOptions, GetPlaygroundMessages, and AskCOTRoom / COTRoomResult via @semoss/sdk's runPixel. Do not use for LLM completions (see model) or document/vector ingestion (see vector).

Use when writing code in an app that does semantic search, RAG, or ingests documents into a vector database on the platform — running nearest-neighbor queries, listing/adding/removing documents, or feeding retrieved chunks into an LLM. Covers VectorDatabaseQuery(), ListDocumentsInVectorDatabase(), CreateEmbeddingsFromDocuments(), CreateEmbeddingsFromVectorCSVFile(), and RemoveDocumentFromVectorDatabase() pixel commands via @semoss/sdk's runPixel, plus listing engines with MyEngines(engineTypes=["VECTOR"]). Do not use for raw SQL/graph queries (see database-engine) or direct LLM calls without retrieval (see model-engine).