Coding agents and review — page 12

Cache-aware, TypeSafe-driven model tier router for Claude Code

TypeSafe's Jev decision model in Claude Code. Ask in plain language, get a typed answer with a calibrated probability.

TypeSafe jev でエージェントの応答を審査し、形式的な完了を Stop hook で差し戻す Claude Code / Codex plugin

Claude Code skill: swap System 2 LLM pipeline components for System 1 TypeSafe Jev decisions via investigation, live three-arm eval, fallback, and an independent judge

Which shell commands may your coding agent run without asking? An allowlist proves what it can; Jev, a no-text model, judges only the rest. Claude Code hook + CLI, measured.

Claude Code Router

Claude Code hook: classify every tool call with TypeSafe's Jev (allow / ask / deny) so Claude never spends tokens judging its own tool calls

Agent plugin: TypeSafe Jev browser automation skills for Grok Bot / Cursor

Agent Skills-compatible skill: govern a Raindrop.io library's collection structure via raindrop-mcp — audit, confirmed restructuring & misplaced-bookmark relocation. Works with Claude Code, Codex CLI, Gemini CLI, GitHub Copilot, Cursor & WorkBuddy.

An LLM router that uses Jev (TypeSafe System One) to dispatch each request to the cheapest model and lowest reasoning effort that will do the job. OpenAI-compatible proxy, Claude Code and Codex wrappers, TypeScript and C#.

Benchmark-style hallucination scoring for Claude Code sessions: atomic claims, fresh-context grader, SimpleQA / AA-Omniscience scoring, per-session tracker.

Jev4JobHunter — AI job application plugin with TypeSafe Jev for job matching. Screen BOSS Zhipin jobs, apply and follow up with Codex / Claude Code. 用 Jev 筛岗位，让 Agent 完成投递与跟进。

A Pi coding-agent extension that routes each prompt to the best backend model — cheap, fast, balanced, or powerful — using local heuristics, with optional TypeSafe Jev classification.

An autonomous product loop for Claude Code that improves a real product overnight: a value gate before any code, an independent validator after, a deterministic driver that knows when to stop, and a one-command installer. Hardened by live runs on an iOS app and its backend.

My pi coding agent extension pack — one-command setup via install.sh

Backup mirror (primary: Cursor Origin)

Equip Claude Code with pi subagents

Shared memory layer for coding agents across harnesses (Claude Code / Codex / Cursor / Kiro), exposed via MCP

Universal plugins for coding agents — Claude Code today, Codex next. Currently: calendar-agenda.

Terminal coding agent — the human harness for the nacelle SDK

Spec-as-truth development for Claude Code: the spec is the source of truth, the code is the artifact. Six skills plus a zero-dependency registry that pins specs by content hash.

Code Review helper — a local app designed to create better, more thorough code reviews. Includes a more-portable Cursor skill with many of the same principles

coding agent that solves gh issues in an isolated checkout, validates changes with deterministic checks, and uses an independent reviewer before creating a draft PR.

Loopback gateway for Claude Code or OpenCode: automatic Anthropic → GLM → DeepSeek failover, breaker-driven, plus optional per-call model routing by Jev (TypeSafe System One) — capability-first across all three providers — with manual pinning, control UI and CLI