Delegates Jac function bodies to LLM calls with typed structured outputs.
npx skills add https://github.com/PMN123/trapdoor --skill jac-by-llm-pmn123
Please help me install this Agent Skill. Skill: jac-by-llm Source: https://github.com/PMN123/trapdoor/tree/main/.agents/skills/jac-by-llm Command: npx skills add https://github.com/PMN123/trapdoor --skill jac-by-llm-pmn123
by llm(...) syntax that replaces a function body with an LLM call returning typed objects, enums, or lists. ## Core Features & Use Cases - Structured Outputs: Return typed objs, enums, lists, and optional values directly from LLM calls, with sem statements serving as the prompt schema. - Tool Use & ReAct Agents: Pass function references via tools=[...] to enable automatic ReAct loops, including bound object methods as tools. - Provider Configuration: Configure OpenAI, Anthropic, Google, Ollama, or built-in local models via Model, jac.toml, or environment variables. - MockLLM Testing: Test LLM functions without API keys using sequential mock outputs. - Use Case: Build a ticket classifier that returns a typed Priority enum, or a ReAct agent that answers questions by calling a word_count tool, all testable offline with MockLLM. ## Quick Start Show me how to write a Jac function that summarizes text into a structured object using by llm with sem descriptions.High-intent search queries and answers about installing and using this skill.
Declare a function with a typed return annotation and replace its body with `by llm(...)`. Define the return type as an `obj`, enum, or list, and describe every field with `sem` statements, which form the prompt schema the LLM sees.
Pass function references in the `tools=[...]` option of `by llm`, which automatically enables the ReAct loop. Each tool needs its own `sem` and per-argument `sem` so the LLM knows when to call it; bound object methods also work as tools.
byLLM supports OpenAI, Anthropic, Google, Ollama, and built-in local models. Configure them via a `Model` glob, `jac.toml` under `[byllm.model]`, or environment variables like `OPENAI_API_KEY` and `BYLLM_DEFAULT_MODEL`.
Use MockLLM from `jaclang.byllm.lib` with a `config` containing an `outputs` list. Mock outputs are consumed sequentially, one per `by` call, and typed returns accept pre-built instances like enum members or object lists.
Inline expressions like `x = "prompt" by llm;` are not supported and raise `NotImplementedError` at runtime despite passing `jac check`. Always declare a named function with `by llm(...)` and call it instead.
Yes, using the `Image` and `Video` types from `jaclang.byllm.lib` as function parameters with a vision-capable model. Image accepts file paths, URLs, bytes, and PIL images; Video requires the `byllm[video]` extra and an fps sampling rate.
nihalnihalaniDelegate Jac function bodies to LLM calls with typed returns, tools, and sem prompts.
nickth3manBuild graph-based LLM workflows with PocketFlow's Node and Flow abstractions.
Unified-Field-Theory-ResearchConfigure runtime-specific model IDs for tier-1, tier-2, and tier-3.
kreuzberg-devCustomize HTML-to-Markdown conversion with visitor hooks for HTML node types.