Matt Pocock Skills v1.2: 3 Releases, New Skills – What to Update in Your Workflow?

v1.2 release overview

Matt Pocock Skills v1.2.0 was released on 2026‑08‑05, followed by quick patches 1.2.2 and 1.2.3. The repository has 213.4k stars. Four human‑machine boundary shifts are introduced: wizard , to‑questionnaire , wait‑what , and prototype .

wizard – automating human‑only steps

Wizard moved from the in‑progress bucket to the Engineering bucket and became model‑invoked . The agent automatically triggers wizard when it detects a step that only a human can perform (provisioning infrastructure, setting credentials, navigating an unfamiliar dashboard, or running a one‑off migration). Wizard generates an interactive bash script from template.sh. The template defines markers such as STAGES, banner, stage, say, open_url, pause, ask_secret, write_env, set_secret, and finish. The script runs ephemerally, writes secrets to .env or GitHub secrets, and never sends secret values to the model.

TOTAL_STAGES=3

stage "打开 Stripe 控制台，导出 restricted key"
  open_url "https://dashboard.stripe.com/apikeys"
  say "登录后进入 API keys，点 Create restricted key，勾选 charges 权限"
  pause "弄好了按回车继续"

stage "把 key 写入本地 .env"
  STRIPE_KEY=$(ask_secret "粘贴 restricted key（输入不回显）")
  write_env ".env" "STRIPE_RESTRICTED_KEY" "$STRIPE_KEY"

stage "同步到 GitHub Actions secrets"
  set_secret "STRIPE_RESTRICTED_KEY"

After v1.2.3 the stage marker accepts only a name; time estimates are removed and progress is counted by the number of stages.

Wizard reads values from .env*, docker-compose*, framework config files, and .github/workflows/* where secrets.* or vars.* are referenced. The generated script can be inspected with bash -n, shellcheck, and value‑flow checks. By default the script is stored under scripts/ and deleted after one run; it can be committed for repeatable setups.

to‑questionnaire – extracting knowledge from others

to‑questionnaire graduates to the Productivity bucket. It creates a Markdown file named to-questionnaire-<slug>.md with a fixed structure: Purpose, From‑To‑How, Context, How to answer, and a list of questions ordered by importance. The core principle is “grill the send, not the subject”: identify the recipient, the decision needed, and then ask targeted questions.

**to-questionnaire: onboarding 评审**

**收件人和用途**
- 收件人：后端负责人（做过 3 次 onboarding 重构）
- 用途：决定是否立项 onboarding 可观测性改造
- 回答方式：异步填写，约 10 分钟

**问题（按重要度排序）**
1. 新员工第几天能独立跑通本地环境？
   （答案留空，给收件人填）
2. 上一次 onboarding 卡住的是哪个环节？
   （为什么问：我们怀疑是环境配置，需要你的数据佐证）

to‑questionnaire complements the /grill‑me skill, which extracts the author’s own knowledge. It is described as a patch for the fact that agents are still hard to collaborate around, pulling the human back into the workflow via a structured questionnaire.

wait‑what – seven‑word re‑pitch

wait‑what is a user‑invoked skill (seven‑line frontmatter). It is triggered when the listener signals “I don’t understand.” The skill asks the model to re‑pitch the last message using Simplified Technical English (ASD‑STE100) and the ubiquitous language defined in CONTEXT.md, rather than compressing it with /tldr or /no‑fluff.

---
name: wait-what
description: Stop. That last message did not land — re-pitch it.
disable-model-invocation: true
---
Wait — I don't understand where you've got to here. Re-pitch that: give me a little bit of context, talk in ASD-STE100 Simplified Technical English, and use the ubiquitous language from `CONTEXT.md`.

The design emphasizes the listener’s state, not the output form. It fixes a failure mode where concision skills become verbose; wait‑what performs a concise, domain‑aligned restatement. The cure for jargon is to build a shared language upfront with /grill‑with‑docs.

prototype – redefining throwaway

In v1.1 throwaway code was deleted after use. In v1.2 prototype becomes a primary source that is archived. Demos are now single‑file HTML pages with interactive tabs, allowing non‑developers to explore scenarios without a build step. After validation the prototype is committed to a prototype/<name> branch (a throwaway branch). The decision (verdict + question) is recorded in an issue/ADR/commit. The archived HTML is a runnable evidence artifact, not production‑grade code.

Prototype HTML contains a state panel, free‑play buttons, and tabbed guided walkthroughs. URL parameter ?variant= selects up to five variants; the winning variant is folded into real code, while losers remain on the throwaway branch. Archiving does not imply production readiness; the prototype remains untested and intended only for evidence.

Dual‑platform installation

v1.2 can be installed via two mutually exclusive routes:

Claude Code plugin : claude plugins install mattpocock-skills. The plugin lists 24 promoted skills in .claude-plugin/plugin.json, updates automatically, and follows a subscription model.

skills.sh : npx skills@latest add mattpocock/skills. This copies files into the project, makes them editable, and requires manual updates with npx skills update.

Codex has no native plugin; its metadata lives in agents/openai.yaml and symlinks to CLAUDE.md. Invocation rules are strict: user‑invoked skills require explicit calls ( disable-model-invocation: true for Claude, policy.allow_implicit_invocation: false for Codex). A user‑invoked skill may call a model‑invoked skill but never another user‑invoked skill.

Minor patches 1.2.2‑1.2.3

#766 (1.2.2): Restores model‑invocation for writing-for-agents in Codex.

#779 (1.2.3): Adds a Redact section to diagnosing‑bugs to mask sensitive data.

#781 (1.2.3): Removes Claude‑specific tool names from several skills to make them harness‑agnostic.

#783 (1.2.3): Simplifies wizard stages to name‑only counting.

Adoption guidance

Claude Code users : install the plugin and enable wizard and prototype for automatic updates and human‑step automation.

Codex users : use skills.sh, adopt wait‑what, and include openai.yaml to understand cross‑harness behavior.

If only human‑step automation is needed, focus on wizard; if domain vocabulary capture is desired, combine wait‑what with CONTEXT.md.

The release does not promise speed gains; its value lies in shifting boundaries—automating human steps, extracting external knowledge, concise re‑pitching, and preserving prototype artifacts.