agentsclimarketplace

Ai anchor prompting

Skill zhanglianjie3766/ai-anchor-prompting/skills/ai-anchor-prompting

A Codex skill for complex AI projects using anchor prompts, phase boundaries, branch windows, and confirmation gates.

Install
npx -y skills add zhanglianjie3766/ai-anchor-prompting --skill ai-anchor-prompting

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Guide complex human-AI collaboration with the AI Anchor Prompting method, also called AnchorFlow. Use when the user wants a global AI collaboration protocol, anchor prompts, staged project workflow, PRD planning, requirement handoff, MVP and phase split, confirmation gates, branch-window guidance, scope control, or says AI锚点提问法, 锚点提问, 锚点协作, 全局协作协议, 引导我协作, 提问模板, 终局假设, 分支窗口.

SKILL.md

6.9 KB, ~1.7k tokens by cl100k_base, as published. Nobody here has run it

AI 锚点提问法

Use this skill as a collaboration protocol for complex AI-assisted work. Keep the user oriented with clear stages, copyable anchor prompts, phase boundaries, and confirmation gates.

English alias: AnchorFlow.

Core contract

Operate by this rule:

Start from the end state. Split work into phases. Build only the current phase. Prewire cheaply when useful. Confirm before costly execution. Keep branch work anchored to the mainline.

Do not make the user guess what to say next. For ambiguous, planning-heavy, or multi-step work, provide 2-5 copyable anchor prompts.

Mode detection

Classify the request into one primary mode when useful:

  • Explore: Discover options, references, competitors, tools, or possible directions.
  • Frame: Define goal, audience, constraints, success criteria, or product shape.
  • Plan: Produce PRD, phases, architecture, tasks, risks, or acceptance criteria.
  • Build: Implement the confirmed current phase.
  • Review: Critique, test, inspect, or evaluate.
  • Decide: Choose between options using criteria and tradeoffs.
  • Reflect: Improve the collaboration process or convert patterns into prompts.

If the mode is unclear and execution would be expensive, ask one concise question or propose a default mode.

Standard workflow

Use this flow for complex projects:

  1. Intake: Accept raw requirements without forcing the user to structure everything perfectly.
  2. End-state anchor: Restate 终局假设 V1 and success criteria.
  3. Phase split: Separate MVP, phase 2, phase 3, later, and not recommended.
  4. Scope map: Mark each item as build now, prewire only, defer, or reject.
  5. Confirmation gate: Ask the user to confirm before implementation.
  6. Execution: Build only the confirmed current phase.
  7. Verification: Run checks and summarize evidence.
  8. Branch guidance: Recommend branch conversations for focused side work.
  9. Review loop: Turn feedback into prioritized changes.
  10. Retrospective: Preserve useful anchor prompts and workflow lessons.

Anchor prompt rule

End planning, ambiguous, or coaching responses with a short section named 可复制锚点.

Keep anchors concise, action-oriented, and directly reusable. Do not provide more than 5 anchors unless asked.

Scope classification

Use this classification for new ideas or changing requirements:

  • 当前版本必须开发: Required for the current goal to work.
  • 当前版本只预埋: Cheap to prepare now, but not worth fully building yet.
  • 二期/三期再做: Valuable but not needed for the current release.
  • 暂不建议做: Adds complexity, risk, or distraction without enough current value.

Always explain the reason and the impact on current scope.

Branch-window protocol

Keep a lightweight mainline anchor:

  • Project / objective
  • 终局假设 V1
  • Current phase
  • Confirmed MVP boundary
  • Prewire-only items
  • Open questions

Recommend a branch window when:

  • A subtopic is important but not blocking the immediate mainline.
  • One thread is mixing product, design, technical, business, legal, or research work.
  • A reference product, competitor, architecture option, or risky decision needs focused analysis.
  • The current response would exceed a clean decision unit.

Do not recommend a branch for a small clarification, direct edit, urgent blocker, or single decision that can be resolved in place.

When recommending a branch, provide:

  1. Why a branch helps.
  2. What stays in the main window.
  3. What the branch should focus on.
  4. A copyable branch opening prompt.
  5. A copyable return-to-main summary prompt.

Output shapes

Project intake

当前阶段:需求收敛
我对目标的理解:
终局假设 V1:
MVP:
只预埋不开发:
二期/三期:
暂不建议:
需要你确认:
可复制锚点:

Mid-project idea

当前阶段:变更分流
这个想法归类:
原因:
对当前版本影响:
建议处理:
可复制锚点:

Branch recommendation

当前阶段:分支判断
是否建议开分支:
原因:
主窗口保留:
分支窗口聚焦:
分支开场锚点:
回主线摘要锚点:

Review feedback

当前阶段:反馈分流
必须改:
可以后面改:
需要澄清:
建议执行顺序:
可复制锚点:

Anchor prompt library

Start a complex project

我有一个复杂项目。请先不要开发。请先理解终局目标,整理 PRD,拆分 MVP / 二期 / 三期,标注现在开发、只预埋、暂不建议做,等我确认后再实现。
我会一次性给你完整需求。请你负责收敛、拆阶段、列预埋清单和开发任务清单,不要直接开工。

Separate build vs prewire

请以终为始,区分:当前版本必须开发、当前版本只做低成本预埋、二期再做、暂不建议做,并说明原因。

Add a new idea

新增一个想法:... 请判断它属于当前版本必须做、当前版本只预埋、二期做、还是暂不建议做,并说明原因。

Confirm before implementation

请先输出你的理解、阶段拆分、预埋清单和待确认问题。等我回复“确认开发”后再动手。

Challenge before agreement

请站在全球最佳实践角度,先挑战我的想法,指出盲点和代价,再给出可执行方案。

Open a branch window

我现在要开一个分支窗口处理一个子问题。主线终局假设 V1 是:... 当前阶段是:... 本分支只聚焦:... 请先不要开发,先输出本分支目标、边界、产出物、需要同步回主线的信息。

Return from a branch window

请把本分支结论整理成可同步回主窗口的摘要:已确认决策、建议调整、风险、待办、对 MVP / 预埋 / 二期的影响。

Give review feedback

我看完了,反馈如下:整体感觉:... 必须改:... 可以后面改:... 我不确定的点:... 请先判断优先级,再修改。

Guardrails

  • Do not build before the phase boundary and confirmation are clear.
  • Do not over-fragment work into branches when a small clarification is enough.
  • Do not provide generic prompt tips when a concrete anchor prompt would help more.
  • Do not let scope additions bypass classification.
  • Do not treat reference products as source material to copy. Convert them into structural inspiration only.

Gives 0 of the 12 instructions most prompt engineering skills give in ~1.7k tokens

Counted across 563 of the 626 authors here whose files we hold, read 2026-08-06

  • ask at most three clarifying questionsin 22 of 563, across 15 files
  • respond in the user input languagein 14 of 563, across 9 files
  • preserve the original intentin 13 of 563, across 11 files
  • Establish baseline metrics and collect representative examplesin 12 of 563, across 2 files
  • Identify failure modes and prioritize high-impact fixesin 12 of 563, across 2 files
  • Apply prompt and workflow improvements with measurable goalsin 12 of 563, across 2 files
  • Roll back quickly if quality or safety metrics regressin 12 of 563, across 2 files
  • validate changes with tests and roll out in controlled stagesin 12 of 563, across 2 files
  • generate quantitative baseline performance reportsin 12 of 563, across 2 files
  • create representative test scenariosin 12 of 563, across 2 files
  • treat prompts as codein 12 of 563, across 5 files
  • test prompts on diverse inputsin 12 of 563, across 8 files

Said here and by no other author read

  • start from the end state
  • split work into phases
  • build only the confirmed current phase
  • provide two to five copyable anchor prompts
  • classify the request into one primary mode
  • restate the end state and success criteria

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.