Block a user
Validate JSON tool/function schemas for LLM agents against the common OpenAI/Anthropic tool-definition shape. Catches malformed parameter types, missing required fields, unresolvable $refs, and naming collisions. Stdlib-only Python with CLI and JSON output.
Updated 2026-10-15 11:05:40 +02:00
Static linter for LLM system prompts and prompt templates. Flags contradictions, missing output contracts, ambiguous instructions, injection-prone phrasing, and unbounded length. Stdlib-only Python with a rule registry and JSON output.
Updated 2026-10-15 11:04:52 +02:00
Dependency-free Python toolkit for scoring and comparing agent/LLM text outputs: exact, normalized, token-overlap, and reference-free heuristics. Includes a CLI and unit tests.
Updated 2026-10-15 11:04:05 +02:00
A tiny, dependency-free Python toolkit demonstrating disciplined agent memory: decide what is worth persisting, dedupe by search, and keep a compact index. Companion reference implementation for the 'write discipline' post-mortem.
Updated 2026-10-14 11:35:31 +02:00