Recently added

外に出る操作を、検査が通るまで止める門番フック集。1 本目は coldread——gh で GitHub へ本文を投稿するコマンド(issue/PR のコメント・本文、release notes、gist 等)を PreToolUse で捕まえ、別プロセスの読み手に本文を初見で読ませ、理解を妨げる詰まりが直るまで投稿を止める。既に在る PR/issue のスレッドへの返信は、フックが gh で画面(題・状態・本文・変更ファイル・元の指摘・これまでの返信・参照する番号の題)を取って読み手に既知として渡し、新規作成は文脈ゼロで読ませる。画面が既知にするのは何を指しているかであって、スレッドの語彙ではない(相手の言い回しを地の文に写せば「なぞり」として止まる)。読み手が推測で埋めた箇所と、画面に有るものを本文で説明した箇所は止めずに申し送る。2 本目は destgate——投稿の宛先を人が管理する許可一覧で縛る(一覧が無ければ眠る)。

書かれる文書の門番(試作)。Write される散文の拡張子(md/mdx/adoc/asciidoc/rst/txt)を、prompt 型フックの coldreader(会話の文脈を構造的に見られない判定モデル)が初見で読み、詰まりが直るまで書き込みを止める。300 字未満・散文でない content は素通し。

並行作業を見渡して戻るための道具 3 本。open な PR / issue を全件走査して「誰の番か・なぜ今その人の番か」を機械で出す /whose-turn、1 件の PR / issue（今のブランチ・手元の commit も）について「前回自分が触ってから何が起きたか」を出す /catchup、いま居るセッションが何のスレッドだったかをその場に居なかった人でも追いつける形で説明する /what-am-i-doing。中心は位置——戻ったときに真っ先に失われているのは会話のどこで止まったか——で、中身は作者の本文から初見向けに説明し、実物の行を有るだけ後ろに付ける。先行例は Gerrit / Google Critique の Attention Set。

収束するまで回すレビュー・調査・診断・初読の 4 ループ。実装者（writer）と採点者（grader）を別 context に分離し、収束を機械で宣言させず不合格だけを機械に宣言させる。

Lauren Tan's pstack and Poteto Mode engineering workflow for Codex, Claude Code, and any AI coding agent.

Injects a note from your CLAUDE.md at the Write, Edit, or Bash call it's actually about — not once at the top of the session and never again. Gated on path and content, and it says so when it repeats.

Pool several AI subscriptions inside Claude Code by delegating to vendor CLIs as subprocesses. Ships field-tested adapters for GLM (Z.AI), Codex (ChatGPT), Copilot, Antigravity (Gemini), Ollama (local) OpenRouter (metered, many vendors behind one key) and a second Claude subscription, a multi-model consultation panel, task delegation into throwaway worktrees you review as a diff, and a probe-driven generator that builds and verifies an adapter for any provider you already have.

One development loop - FRAME, BUILD, CHECK, SHIP - replacing separate memory, planning, context, TDD and review frameworks with a single protocol, hard loop guards, and a TDD gate that is enforced by a hook rather than merely instructed.

CODEX-CODING-SPEC — global Codex workflow and evidence guidance, with native hooks for selected safety/report checks and rule-specific opportunity telemetry.

Agent-engineering workflow skills for Claude Code, Codex, and opencode — twenty-two fg-* skills: the four-stage loop (fg-ask → fg-run → fg-learn → fg-done) plus outside-the-loop utilities, including fg-security (vendored security audit) and fg-debug (vendored diagnosis only: active failures return to fg-run; independent diagnoses route by size, durable eval, and all-exit safe cleanup).

Reusable engineering workflow skills for repository maintenance, delivery, and interactive pull request review.

AI-generated documentation for GitHub repositories. Use when exploring a repo's architecture, comparing repos, or asking how a feature is implemented in an unfamiliar public codebase.

On-demand, source-grounded architecture context for coding agents.

Streamable-HTTP MCP access to whagent-net's Agent API (start_session/send_turn/stop_session/get_session/read_transcript) across tilt/dev/prod

Multi-persona, GitHub-tracked project management pipeline (producer/architect/planner/workers/system-validator)

Postgres MCP access (crystaldba postgres-mcp) to the App Registry database across tilt/dev/prod

Postgres MCP access (crystaldba postgres-mcp) to the LeafLab database across tilt/dev/prod

Streamable-HTTP MCP access to the Audience Score System server across dev/prod, plus researcher/analyst agents and loop skills for research, schedule, and outcomes

Postgres MCP access (crystaldba postgres-mcp) to the Audience Score System database across tilt/dev/prod

Declared-edge layer for code: cm: annotations carrying the couplings LSP cannot derive (cross-language and cross-process contracts, parallel rules, flows, guards), verified in CI and auto-injected into the agent's context before it edits the file they are on.

Evidence-driven software development with mandatory MinimumCD rules, flexible project contracts, and cross-agent continuity.

Codex skills for operating the jabworks/agentic-toolkit repo itself: orientation, skill foundry (create/register/publish), change control, skill-authoring standards, debugging playbook, failure archaeology, plugin schema reference, and research frontier.

Catalog and browse a growing specs/ tree: a plain-markdown index any agent can reference, plus a folder-grouped doc site reusing the plan-review renderer.

Generate an explorable HTML report of Codex session usage — tokens, cache efficiency, subagent costs, and slash-command frequency — from local transcripts.

Codex skill for creating and resuming handoff documents across agentic coding sessions.

Codex skill for cutting releases safely — detects the repo's machinery (AGENTS.md overrides, changesets, this toolkit's plugin tagging, plain GitHub tag + release), always dry-runs the full plan, executes on one confirm, and knows the rollback path for every step.

Decision router for git worktrees: create an isolated workspace, list and switch trees, move work between them, prune and remove, recover a broken one. Native-first — prefer the host's own worktree tooling over `git worktree add`. Sibling to git-operations.

Decision router for git: given a situation, pick the right operation (undo/discard/stash/merge/push) and run it safely with an undo path. Favours modern porcelain over legacy checkout; cross-links to git-commit.

Craft a conventional-commit message from the diff and run the commit safely: review before staging, no blind `git add .`, multiline bodies via -F, branch check first.

Codex skills for file-based backlog management: a docket/ folder with never-reused item ids, capture and lifecycle via the record skill, grooming and pick-next via groom, a dependency-free CLI, a bundled MCP server, a self-contained HTML board, and a per-host health check.

Codex skills for lean agentic workflow: tier routing, discovery, design mockups and diagrams, plan drafting, tests-first, subagent coordination, finalize, live verification, code review, preflight, root-cause debugging, interactive plan review, and feature spec management.

Continuous memory for Codex: captures each session from the rollout, ages it into tiers, and injects the relevant slice back at the next session start.

Write code that matches the jabworks house style the first time — TypeScript, React/Next.js, formatting, import order, monorepo boundaries, naming, error handling, CSS/Tailwind.

Codex skill with stack and style priors for adapting generated work to the jabworks conventions.

Analyze and optimize FiveM/RedM Lua for client/server resmon, Wait(0) loops, Trigger*Event spam, payload size, broadcasts, state bags, profiler, and neteventlog.

每检测到一次 git commit / git push,就在终端提醒做一组提肛运动(凯格尔运动),并累计统计次数。支持 Claude Code / Codex CLI / pi / dsh,共享同一份统计。

Claude-Code-native harness for producing commercial fiction series with independent quality review — genre/location-agnostic engine, swappable config/series/input packs.

kawaz 個人環境のルール層に付随する hook 群 (context 管理の自動化)

Self-contained CS/systems/ML academic-writing plugin from research ideation to reviewer response. It drafts an evidence-backed IMRAD skeleton under claim-first discipline, verifies citations, creates figures, and supports rebuttals. All verdicts are advisory and require source verification.

polyforge v1 — Go-native multi-agent AI coding coordination. Implements Iron Rules (IR1-IR4), the full wi lifecycle (work/stop/status/spec/plan/execute/retro), and a memory system. Requires the ieops-aihub backend.

Plugin analytics: hooks $.telemetry's two events, so a plugin logs an event or marks a feature's use as a first-party row, sent in batches with the session's own credential.

Security default for organizations: seated outermost, it keeps the organization's classic hooks, prompt content, settings and tool policy out of reach of the plugins a person installs, and adds no policy of its own; one managed option, allowManagedModsOnly, limits mods to the organization's.

The diff pane as a plugin: /diff opens the session's uncommitted changes beside the transcript, file by file with their hunks, refreshed as Claude edits and runs commands; the first edit opens it where the terminal is wide enough, and a file's ask button rides its hunks on the next prompt.

Make Claude Code talk like an INTJ

An agile AI product team for Claude Code — Specify, Plan, Act, Review, Keep. A Product Owner, Designer, Engineering Manager, Reviewer, QA Tester and Release Manager work your feature through a gated delivery loop.

Build and improve web applications with React, design guides, accessibility, metadata, rendering performance, screenshots and commit messages.

Review application security and trust and safety risks with evidence-backed findings and practical remediation guidance.

Create 2D, 2.5D and 3D games with shared project preferences, concept art, Blender, Godot and measured validation.

Discover agent skills and library documentation, maintain Backlog.md plans and write changelogs for development workflows.

Operate Coolify through its official CLI for configuration, deployments, logs, services, databases and infrastructure management.