Scoring model and reproducible benchmark for agent-generated Keboola UI — turns @keboola/validate-ui verdicts into a tunable 0-100 BenchScore and runs a fixed fixture target set to record a committed baseline

J-Rig seven-layer binary eval CLI for Claude Skills — the `j-rig` command: package integrity, trigger/functional/regression/baseline scoring, optimizer, and rollout-gate evidence. Self-contained (bundles the internal eval engine).

Skill Refiner pure core: bounded-edit apply transform, deterministic synthetic eval-set bootstrap, the Pareto-dominant acceptance gate (DR-028 P0-RATIFY-1), and the swappable RefinerStrategy interface (AC-13).

Canonical contracts kernel for the Intent Eval Platform — TypeScript types, JSON Schemas, Zod validators, and state machines for the 16 canonical entities.

Mathematical expression evaluator

LLM prompt eval gate for CI: run YAML-defined eval cases against your prompts on every PR, post a pass/fail report, and fail the build when outputs regress.

ICE(Interactive Connectivity Establishment) Implementation for TypeScript (Node.js)

The agentic meta-harness — freeze the model, evolve the harness. An open runtime that routes each query to the cost-optimal model, evolves its own harness (planner/context/reviewer/retry/tool/memory/score policy) and autonomously repairs code, then orches

Adds support for environments that disallow support of new Function

TypeScript execution environment and REPL for node.js, with source map support

decycle your json

Parse Content Security Policy directives.

Evaluate and improve AI agents from runs, traces, judges, and feedback. Compare candidates, cluster failures, measure lift, and gate releases.

Safer version of eval()

Build and evaluate JavaScript strings safely via tagged template literals

Freeze the model, evolve the harness. Two measured applications: (1) SWE-bench code-repair — conformant GLM->Opus empty-patch cascade resolves 51.3% Lite (n=300) and 55.6% Verified (278/500, Wilson 95% CI [51.2, 59.9], official swebench gold eval, no gold

Atlassians wrapper for the Statsig js-lite client.

The Dak Stack deepen generator — the self-improving loop's ORIGINATOR (the eval gate is the validator; this proposes what the gate judges). proposeRevision is a GEPA/DSPy-style hill-climb over a skill's natural-language scaffolding (the SKILL.md body): a

OCI NodeJS client for Golden Gate Service

Evaluate mathematical expression with the built-in math object