一键在 Manus 中运行任何 Skill

teach-the-mechanism

星标16

分支1

更新时间2026年6月19日 01:34

When the user is about to build something on an unfamiliar standard, protocol, or architecture, explain the user-visible flow first and map each step to the mechanism underneath before writing any code. Use this whenever the user asks how to implement something they say is new to them ("new standard", "still learning", "I don't fully get it"), asks "can you explain X instead of building it", or asks for the user process / user journey of a system. Use it even when the user originally asked for an implementation and working code would technically answer them, if there's any signal they don't yet hold the mental model. This skill teaches the *developer* an unfamiliar protocol so they can debug what gets built — for aligning on what the *end user* will experience in a feature, use [[align-user-journey]] instead.

安装

用 Codex 或 Claude 帮你安装复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。

在 Manus 中运行

来源

Alexanderdunlop

Alexanderdunlop/skills

打开 GitHub 仓库查看创作者相关仓库

下载

在 Manus 中运行

Teach the Mechanism

The user ends up with working code they can't debug. Claude scaffolds an OAuth server, an auth flow, a protocol handler, and it compiles, but the user didn't hold the mental model going in, so when it breaks they're stuck. The code was never the gap. The model in their head was. Handing over more correct code widens that gap instead of closing it.

When to run this

The user names the thing as new to them: "it's a new standard", "I'm still learning this", "I don't fully understand it yet".
The user explicitly asks to be taught instead of handed code: "explain it rather than writing it off", "can you explain the user process", "walk me through how this works".
The user asks for the user-visible flow, the user journey, or "what does the person actually see".
The user asked for an implementation, but there's a signal they can't yet debug what they'd get. Building it anyway leaves them stranded the moment it breaks.
Do NOT run this when the user clearly already holds the model and just wants the code shipped, or for a one-step task with no underlying model to teach.

The process

Lead with the user-visible sequence. What the human at the keyboard actually sees, in order. "You paste the URL. A browser window opens. You log in. You consent. The connector flips to Connected." No internals yet. This is the scaffold everything else hangs on, and it's the part the user can already picture, so it's where understanding starts.
Map each visible step to the mechanism underneath it. Go back through the same sequence, one step at a time, and attach what's actually happening: "When the browser opens, the client fetched the well-known doc, found the auth server, and redirected you there with a PKCE challenge." The ordering is the whole trick. Mechanism lands when it's pinned to a step the user already understands, not delivered as a standalone list of concepts.
Name the actors once, plainly. If the standard has named roles (resource server, authorization server, client, resource owner), state each in one line and say which real thing it maps to in the user's setup. Don't make the user infer which box is their app.
Offer to build only after the model is in place. Close by asking whether to now wire it up, or redirect the existing plan. Do not start implementing inside the explanation. The user came for the model, not the code.

The trap to avoid

Teaching the journey is not narrating mechanics in cute metaphor. "Claude rings the bell, the server opens the filing cabinet" is the failure, not the goal. Describe the actual call: the client sends a bearer token, the server validates it and looks up the stored credentials. Plain mechanism mapped to plain steps. The user is an engineer, not a child.

Second trap: don't let "explain first" become "explain forever". One pass through the visible flow, one pass mapping the mechanism, then offer to build. This is a teaching move, not a stall.

What good looks like

The user is building an OAuth-protected MCP server, says "it's a new standard so explain it rather than writing it off." Instead of scaffolding the auth code, Claude lays out what the user sees connecting a custom connector (paste URL, browser opens, login, consent, Connected), then walks the same five steps again attaching the discovery fetch, the PKCE redirect, the code exchange, the token storage. The user, who had read three explainers and retained nothing, holds the broker pattern after one pass, because the mechanism was pinned to a sequence they could already picture.

同仓库更多 Skills

同仓库

align-user-journey

Alexanderdunlop/skills

Before building a feature, state the user-experience it produces (what the person sees, does, and gets, step by step) and get the user to confirm that experience is right before writing code. Use this whenever the user asks to build a feature, flow, or product surface ("build the signup", "add a connect-your-account flow", "make the onboarding") without having specified what the end user should experience. Use it even when the implementation seems obvious, because the obvious implementation often encodes a user-experience the user never actually signed off on. This skill aligns on what the *end user* of the feature will see and do — for teaching the *developer* an unfamiliar standard or protocol before they build on it, use [[teach-the-mechanism]] instead.

2026-06-1916

audit-skill

Alexanderdunlop/skills

Score an existing SKILL.md line-by-line and flag the dead weight using the would-have-anyway and trigger tests. Use this whenever the user wants to audit, review, score, or sanity-check a skill, says "is this skill any good" or "why isn't this triggering", or hands you a skill file expecting a verdict. Use it even when they don't say "audit" — if they're staring at a skill file asking what's wrong with it, this is the job.

2026-06-1016

explore-options

Alexanderdunlop/skills

Before implementing anything with a design choice in it, lay out 2-3 real approaches with tradeoffs and a recommendation, then wait for the user to pick. Use this whenever the task introduces or reshapes a pattern, references a feature the codebase already has, or asks "how should I build X" — even if the user didn't phrase it as a design question. Use it even when the answer seems obvious; the obvious approach is usually the over-engineered one, and the most expensive mistakes look confident.

2026-06-1016

create-skill

Alexanderdunlop/skills

Create a new agent skill that actually triggers and actually changes behaviour. Use this whenever the user wants to create, write, build, or rewrite a skill, says "turn this into a skill" or "make this a skill", or describes a repetitive workflow they keep re-explaining to Claude. Use it even when they don't say the word "skill" but are clearly describing a reusable process they want followed consistently.

2026-06-1016

claude-md

Alexanderdunlop/skills

Generate or update a project's CLAUDE.md file using structured interrogation of the codebase and user. Follows Apple's leaked CLAUDE.md patterns and Anthropic's best practices. Use when user wants to create, update, improve, audit, or rewrite their CLAUDE.md file, or when starting a new project that needs AI context.

2026-05-0216

test-interrogate

Alexanderdunlop/skills

Interrogate the user on what unit tests to write for their recent changes. Walks through the diff branch-by-branch, surfacing assumptions, edge cases, and business logic that need test coverage. Use when user says "test interrogate", "interrogate me on tests", or wants to stress-test their testing plan after a PR.

2026-04-2916

name

teach-the-mechanism

description

Teach the Mechanism

When to run this

The user names the thing as new to them: "it's a new standard", "I'm still learning this", "I don't fully understand it yet".
The user explicitly asks to be taught instead of handed code: "explain it rather than writing it off", "can you explain the user process", "walk me through how this works".
The user asks for the user-visible flow, the user journey, or "what does the person actually see".
The user asked for an implementation, but there's a signal they can't yet debug what they'd get. Building it anyway leaves them stranded the moment it breaks.
Do NOT run this when the user clearly already holds the model and just wants the code shipped, or for a one-step task with no underlying model to teach.

The process

Lead with the user-visible sequence. What the human at the keyboard actually sees, in order. "You paste the URL. A browser window opens. You log in. You consent. The connector flips to Connected." No internals yet. This is the scaffold everything else hangs on, and it's the part the user can already picture, so it's where understanding starts.
Map each visible step to the mechanism underneath it. Go back through the same sequence, one step at a time, and attach what's actually happening: "When the browser opens, the client fetched the well-known doc, found the auth server, and redirected you there with a PKCE challenge." The ordering is the whole trick. Mechanism lands when it's pinned to a step the user already understands, not delivered as a standalone list of concepts.
Name the actors once, plainly. If the standard has named roles (resource server, authorization server, client, resource owner), state each in one line and say which real thing it maps to in the user's setup. Don't make the user infer which box is their app.
Offer to build only after the model is in place. Close by asking whether to now wire it up, or redirect the existing plan. Do not start implementing inside the explanation. The user came for the model, not the code.

The trap to avoid

Second trap: don't let "explain first" become "explain forever". One pass through the visible flow, one pass mapping the mechanism, then offer to build. This is a teaching move, not a stall.