saved
semantic-algos: Why Combinators
Hraness wrote this summary from a saved copy of the source. Quotations are taken word for word from the source.
gist
semantic-algos, also called Why Combinators, is Rob Cheung's standard library of reasoning procedures that compose interpretive operations over questions: tracing causes, testing premises, comparing options, reframing, or giving ideas literary form. Each skill is a repeatable semantic algorithm whose execution still requires judgment. In harnesses that can launch fresh subagents, sem-run compiles a request into an inspectable computation graph, runs one isolated subagent per operation, and returns the run as linked Markdown. There is no parser or executable runtime; the compiler turns prose into more explicit prose.
ideas
- A semantic algorithm is a repeatable interpretive procedure. Control structure, output form, stopping rule, and guardrails matter; execution still requires judgment.
- The pack is a standard library, not a prompt list. Causes, decisions, reframes, and literary forms are operators that can be composed into Sem programs.
- sem-run is the meta-program. It compiles a request, gives each application a fresh no-history subagent, and records the run as linked Markdown files.
- State lives in files, not chat. A run can be inspected or resumed without conversational memory; there is no parser, type checker, or checked-in executable runtime.
- The behavioral test is removal. If dropping the procedure would not change the answer's structure, the skill is only a writing prompt.
quotes
“Stdlib of reasoning procedures to milk the humanities out of LLMs”
“Each procedure is repeatable; its execution requires judgment.”
“Sem has no parser, type checker, generated code, or checked-in executable runtime.”
“If removing the procedure would not change the answer's structure, the skill is only a writing prompt.”