100 LLM workflow tools for prompts, guardrails, RAG, and eval

Toolcore added 100 more client-side helpers for LLM workflow planning—prompt section audits, JSON shape checks, guardrail pattern scans, RAG chunk packing, and local eval metrics. They live under AI tools alongside token, embedding, and agent batches. None call a generative API.

Prompt craft

Count sections, XML/JSON blocks, few-shot examples, and constraint keywords before you send—start with prompt section header count, few-shot block count, and system block token share.

Structured output

Validate JSON shape hints locally—brace balance, required keys, nesting depth, and schema property counts—try JSON brace balance, required keys check, and JSON max nesting depth.

Guardrails scan

Report-only heuristics for email, API keys, injection phrases, and paste risk—see paste risk score, API key pattern scan, and ignore-instructions hint. Pair with paste scan for redaction workflows.

RAG pack

Plan chunk fit, overlap, citations, and context headroom—open chunk pack fit, chunk overlap chars, and context headroom chars.

Eval local

Compare reference vs hypothesis with word overlap, length ratio, JSON compliance, and format scores—try word overlap percent, JSON format compliance, and format compliance score.