dag-typesafe

# dag-typesafe Deterministic type safety via directed acyclic graphs for reasoning language models. Analyze any repo's public API surface, extract a typed node registry, compose validated execution DAGs from natural language, and compile them into native pipeline code with GraphSentry-style `(artifact, certificate)` verification at every node boundary. ## Category Code Scaffolding & Templates ## Core Concepts Three layers compose into a single architecture: 1. **Typed Node Registry** — the repo's public functions/classes extracted as typed nodes with JSON Schema input/output contracts. The LLM selects from this registry; it never generates arbitrary code. 2. **DAG Plan** — a language-neutral execution graph where nodes reference registry entries and edges are schema-validated. No cycles, all inputs satisfied, all types compatible. 3. **Certificates** — each node emits an `(artifact, certificate)` pair. Certificates are deterministic predicates evaluated from logged evidence. Failed certificates halt the pipeline with diagnostic context. Based on GraphSentry (Li et al., 2026). ## Commands Parse `$ARGUMENTS` to determine which command to run: ### `analyze` Extract a typed node registry from the current repository. ```bash python3 ~/.claude/skills/dag-typesafe/scripts/analyze.py [--language python|typescript|auto] [--output dag-registry.json] ``` 1. Detect repo language(s) from file extensions and config files 2. Run the appropriate extractor(s) fro...