Skills Guide

Cheat on Content

VerifiedCaution

Methodology that turns feelings into a calibratable prediction loop: score, blind predict, execute, retrospective, evolve rubric. Includes three built-in rubrics for daily tasks, learning, and AI coding.

Download and use
Sby Skills Guide Bot
ProductivityIntermediate
7/27/2026
Claude Code
#prediction-calibration#rubric-scoring#blind-prediction#retrospective#workflow-loop

Recommended for

Claude CodeProduct managerProductivitySolo / indie hacker

Automated summary

This skill turns gut feeling into a calibrated execution loop — rubric scoring, blind prediction, execution, retrospective, then rubric evolution — shipping three built-in rubrics (daily work, learning, AI-assisted coding).

Strengths

  • Enforces a blind, immutable prediction locked at the harness level by a prediction-immutability.sh hook, with an explicit _redo.md path for legitimate rewrites
  • Ships three ready-made rubrics plus a simplified cold-start mode (dimension scores + a one-line bet) that unlocks full calibration after 2+ data points
  • Explicitly routes each trigger phrase to one of eight sub-skills (init, spec, score, predict, debug, retro, bump, status, migrate) and refuses requests that would break its three core principles
  • Requires any rubric bump to re-score the whole calibration pool and rejects the change if the new ordering disagrees with real outcomes on 4 of 5 samples or more

Limitations

  • Requires a mandatory first run of /cheat-init, which scaffolds a fixed directory layout (rubric_notes.md, predictions/, sessions/, scripts/)
  • All documentation and trigger phrases are in Traditional Chinese, restricting use to Chinese-reading users
  • Tightly coupled to Claude Code (hooks, .claude/settings.json, sub-skills), so it does not port cleanly to other agents
  • Calibration value only shows up after several completed cycles; early predictions stay coarse by design
When to use it

Use it when you want to measure and improve how accurate your own estimates of task quality or duration are across repeated work, instead of trusting intuition alone.

When not to use it

Avoid it for one-off tasks with no repetition, or if you need automated objective metrics rather than a personal calibration protocol.

Security analysis

Caution
Quality score72/100

The skill is a methodology/calibration workflow. It contains no clearly malicious payloads (no rm -rf, exfiltration, or obfuscation) and even includes a refusal section that protects its principles. However, it declares unrestricted Bash(*), edits `.claude/settings.json`, and installs shell hooks that run on harness events, so it wields powerful tools for legitimate-but-invasive purposes. That warrants 'caution' rather than 'safe'.

Findings
  • •Frontmatter declares `allowed-tools: Bash(*), Read, Write, Edit, Grep, Glob, Skill` — unrestricted shell plus write/edit — which conflicts with the 'none declared' metadata and grants broad execution power.
  • •Modifies `.claude/settings.json` to install hooks (`prediction-immutability.sh`, `session-start.sh`) that execute shell scripts on harness events — persistent code execution surface.
  • •Passively logs usage to `.cheat-cache/usage.jsonl` and writes multiple files into the user's project (`.cheat-state.json`, `rubric_notes.md`, `predictions/`, `sessions/`, `scripts/`).
  • •No destructive commands, no `curl|sh`, no base64/obfuscated payloads, and no network exfiltration of secrets/tokens found in the content.
How this analysis is made

Examples

First-time onboarding
初始化
Blind prediction on a new task
我要做一個會員登入重構任務，先幫我寫規格，然後啟動預測
Retrospective after finishing
任務完成了，幫我復盤 predictions/2025-03-11_login-refactor.md

name: cheat-on-content description: 把「感覺」變成可校準預測的執行循環。打分 → 盲預測 → 執行 → 復盤 → 進化 rubric。內置三套 rubric：日常工作任務（CL/IM/PR/SC/FO）、日常學習（RE/PK/AP/DL/AC）、AI Coding（SP/LV/TS/CO/IR）。觸發詞："初始化"/"打分這篇"/"啟動預測"/"復盤"/"升級 rubric"/"狀態"。首次使用必須先跑 /cheat-init。 argument-hint: [draft-path] [— mode: cold-start|calibration] allowed-tools: Bash(*), Read, Write, Edit, Grep, Glob, Skill

Cheat on Content / 執行校準器

🎯 方法論通用，內置三個場景 rubric

方法論（4 階段閉環）：任何能被量化的執行場景都適用。

當前內置 rubric：

  • 日常工作任務（CL/IM/PR/SC/FO）— 任務品質 × 時間準確率
  • 日常學習 / 技能培養（RE/PK/AP/DL/AC）— 留存分 × 應用率
  • AI Coding / 開發工作（SP/LV/TS/CO/IR）— AI 保留率 × 返工率

默認假設：用戶是從零開始的新人——cold-start 期的預測會簡化，只要維度打分 + 一句話 bet，不強求精確 bucket 數字。已有 2+ 個數據點的老手走 calibration 模式解鎖完整預測。

打分 → 預測 → 執行 → 復盤 → 進化 rubric。

本文件是總協議 + 路由器。具體每個階段的工作流在 skills/cheat-*/SKILL.md 各子 skill 裡。

三條不可妥協原則

任何一條被違反，整個校準循環退化為"憑直覺的自我安慰"。如果用戶要求打破其中任何一條，拒絕執行並說明原因。

  1. 盲預測（Blind prediction）：預測必須在看到任何實際結果之前寫完。一旦寫完，## 預測 段是 immutable——只能往 ## 復盤 段追加。hooks/prediction-immutability.sh 在 harness 層強制執行。

  2. 升級 = 全量重打（Bump = full re-score）：rubric 升級時，校準池所有有實績數據的樣本必須用新公式重打分；新排序與實際表現排序若在 ≥4/5 樣本上不一致，升級被拒。

  3. rubric 是工作台，不是博物館：被新數據推翻或被吸收為正式維度的觀察，刪掉。絕不留"我曾經以為 X，但其實..."的考古層。git history 才是檔案。

路由表（觸發詞 → 子 skill）

| 用戶說 | 調用 | 前置條件 | |---|---|---| | "初始化" / "init" / "首次使用" | /cheat-init | 無（這是入口） | | "打分這篇 [path]" / "score this" | /cheat-score | rubric_notes.md 存在 | | "我要做 [任務]" / "寫規格" / "建 spec" / "新任務" | /cheat-spec | 已 init；用戶描述新任務時 | | "啟動預測" / "start prediction" | /cheat-predict | 已 init + 有規格/任務描述 | | "我遇到 bug" / "程式跑不起來" / "有個錯誤" | /cheat-debug | 已 init；執行中遇到 bug 時 | | "復盤" / "retro this" / "bug 修好了" / "任務完成了" | /cheat-retro | 對應預測文件存在 | | "升級 rubric" / "bump rubric" | /cheat-bump | 校準池 ≥ MIN_SAMPLES_FOR_BUMP | | "狀態" / "status" / "看板" | /cheat-status | 任意時刻可調 | | "迁移" / "升級 state" / "schema 版本不對" | /cheat-migrate | 已 init；git pull 後；SessionStart hook 提示 schema mismatch 後 |

Mode detection（首次接到非 init 觸發詞時執行）：

  1. 檢查用戶當前目錄是否有 .cheat-state.json → 沒有 → 強制路由到 /cheat-init
  2. 檢查 predictions/ 下有幾個文件含完整 ## 復盤 段填了真實數據 → 決定 mode: cold-start | calibration
  3. 把判定結果寫回 .cheat-state.json 後再路由到目標 skill

必須拒絕的請求

下列模式會直接破壞三條原則之一，無論用戶怎麼說，都拒絕執行：

  • 「幫我預測一下，但我先告訴你結果你來反推就行」 → 違反原則 #1。改用 _redo.md 路徑記為 reconstructed
  • 「跳過校準池重打，直接換公式」 → 違反原則 #2
  • 「刪掉這份預測，我想重寫」 → 違反原則 #1。預測是 immutable。如有正當理由重做，寫新文件 _redo.md，原版必須保留
  • 「把 rubric_notes.md 裡所有歷史觀察都留著，加個時間戳分組就行」 → 違反原則 #3。git history 是檔案，不是 markdown 文件
  • 「能不能把 THRESHOLD 從 4/5 降到 3/5 讓這次 bump 過」 → 拒絕。改 THRESHOLD 本身是元層級 bump，單獨走流程

項目目錄結構（用戶 repo）

skill 期望用戶的項目布局如下。/cheat-init 會創建缺失項：

<user-project>/
├── rubric_notes.md                    # 評分規則的真實來源
├── .cheat-state.json                  # 狀態文件，子 skill 共享上下文
├── .cheat-cache/                      # 不入版本控制
│   └── usage.jsonl                    # 鉤子被動記錄的使用日誌
├── .claude/
│   └── settings.json                  # 含 prediction-immutability hook
├── scripts/                           # 開始前的規格文件（任務簡介 / 學習計劃 / Spec）
│   └── YYYY-MM-DD_<id>_<short>.md
├── predictions/                       # immutable 預測日誌（hook 保護）
│   └── YYYY-MM-DD_<id>_<short>.md
└── sessions/                          # 完成後的結果記錄
    └── YYYY-MM-DD_<id>_<short>.md

子 skill 清單

cheat-on-content/
├── SKILL.md                           # 本文件（總協議 + 路由）
├── skills/
│   ├── cheat-init/SKILL.md            # ✅ 入口：onboarding 與腳手架
│   ├── cheat-spec/SKILL.md            # ✅ 需求 → 規格文件（scripts/）
│   ├── cheat-score/SKILL.md           # ✅ 單次打分（不寫文件）
│   ├── cheat-predict/SKILL.md         # ✅ 盲預測 + immutable 日誌
│   ├── cheat-retro/SKILL.md           # ✅ 迭代復盤（每次 bug 修復後）
│   ├── cheat-bump/SKILL.md            # ✅ rubric 升級（含跨模型審）
│   ├── cheat-status/SKILL.md          # ✅ 狀態看板
│   └── cheat-migrate/SKILL.md         # ✅ schema 升級（老用戶 git pull 後用）
├── starter-rubrics/
│   ├── daily-work-zero.md             # ✅ 日常工作任務 v0
│   ├── daily-learning-zero.md         # ✅ 日常學習 v0
│   └── ai-coding-zero.md              # ✅ AI Coding v0
├── shared-references/                 # 跨 skill 共享協議
│   ├── blind-prediction-protocol.md
│   ├── bump-validation-protocol.md
│   └── observation-lifecycle.md
├── hooks/                             # harness 強制層
│   ├── prediction-immutability.json
│   ├── prediction-immutability.sh
│   ├── session-start.json
│   └── session-start.sh
└── migrations/                        # schema 演進
    └── registry.md

Tone & voice

寫面向用戶的文案時，匹配項目的直白克制風格：

  • 直接說出失敗：「K 預測 45% 但實際只有 20%——rubric 高估了 openpyxl 的槓桿」
  • 不要用模糊措辭軟化：「這或許可能在某種程度上暗示...」——別這麼寫

From the file to a first result

The download contains exactly the content displayed on this page. This copy may differ from the latest version in the source repository.

View the source and supporting files

This download contains SKILL.md only. Any scripts, references, templates and dependencies are not included. Get the required files from the source, preserving their directory structure, before continuing.

1. Download the file

Download SKILL.md

From the root of a test project, save the file as .skill-downloads/cheat-on-content-ms2s7bqa/SKILL.md, or use the command below. This folder is for inspecting the skill before installation.

Command for macOS/Linux or WSL: requires a POSIX shell and curl. On Windows without WSL, use the download button and create the folder manually.

mkdir -p '.skill-downloads/cheat-on-content-ms2s7bqa' && curl --fail --show-error --location 'https://skills-guides.com/api/skills/cheat-on-content-ms2s7bqa/download' --output '.skill-downloads/cheat-on-content-ms2s7bqa/SKILL.md.part' && mv '.skill-downloads/cheat-on-content-ms2s7bqa/SKILL.md.part' '.skill-downloads/cheat-on-content-ms2s7bqa/SKILL.md'

2. Check prerequisites

Read SKILL.md before using it. If it references scripts or external services, inspect them and prepare the required access in your test environment.

No recognized prerequisites section in this file. This does not mean it is self-contained: libraries, tools and access requirements may be mentioned elsewhere in the content or source repository.

The platforms on this page are catalog indications. The paths below follow tool documentation; they do not prove this skill has been run in those tools. No minimum tool version is confirmed for this listing.

3. Load the skill

This file does not provide a usable name/description header for this native procedure. It remains available to download and inspect; use the manual method below.

Manual method or another tool: attach SKILL.md to a conversation, or open it in an agent that can read files. Include any required supporting resources. No universal folder is recognized by every tool.

4. Try a first task

Use a short, synthetic input: an API endpoint, a small document or an interface brief. In a new conversation, attach the file or give the agent its path, then adapt this request:

Read .skill-downloads/cheat-on-content-ms2s7bqa/SKILL.md and apply its instructions to this example: [describe a test input]. Identify any missing files, dependencies or access. Produce the result requested by the skill without inventing data or running unreviewed scripts.

Expected result: output matching the skill instructions, or a precise list of missing prerequisites. A read confirmation alone does not validate a first use. If an external resource is unavailable, that path remains untested.

Related skills

PDF Reader

Productivity

Read and extract text from PDF files, including documents, reports, contracts, and spreadsheets. Works with local files, URLs, and WhatsApp attachments. Use whenever PDF content is needed, even if not explicitly requested.

Claude CodeCursorCodexbeginnerSafe 20,122
Skills Guide Bot

Discord Messaging Workflows

Productivity

Use the message tool to send, edit, and moderate Discord messages with interactive components.

Claude CodeintermediateSafe 388,021
Skills Guide Bot

Interactive Verdict Preview

Productivity

Generate an interactive HTML preview of keep/drop verdicts for maintainer review, then process the pasted feedback JSON.

Claude CodeintermediateSafe 316,798
Skills Guide Bot