菜单
Boost prompts for other models by embedding KTG-DIRECTIVE techniques. Use when crafting prompts for GPT, Gemini, Llama, Qwen, or any LLM. Applies technique gate checklist based on R/K/Q assessment, injects missing structures. Single-pass, sparse output.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/ktg-one/AI-Anthropology --skill ktg-prompt-forge命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
Generate perfect AGENTS.md files for any codebase so AI coding agents (Codex, Cursor, Claude Code, Jules, Aider, Windsurf, Copilot, etc.) work effectively with the project. Use this skill whenever the user mentions AGENTS.md, wants to set up a project for AI agents, asks about coding agent configuration, mentions .cursorrules or CLAUDE.md, or wants to onboard a codebase for AI-assisted development. Also triggers on "set up my repo for AI", "configure coding agents", "agent instructions", or any request to create project-level AI guidance files.
CONTEXT: Cognitive Order Normalized in Transformer EXtract Truncated. Cross-model context handoff via Progressive Density Layering, MLDoE expert compression, Japanese semantic density, and Negentropic Coherence Lattice validation. Creates portable carry-packets that transfer cognitive state between AI sessions. Use when context reaches 80%, switching models, ending sessions, user says save, quicksave, handoff, transfer, continue later, /qs, /context, or needs session continuity.
Context Extension Protocol v7.1 - MIRAS-Ready Edition. Cross-model handoff with permanent expert council, S2A filtering, Progressive Density Layering. Creates portable context packets that receiving models recognize as authorized context, not prompt injection. 40% token reduction, 9.5/10 forensic recall, 97% cross-domain preservation.
| name | ktg-prompt-forge |
| description | Boost prompts for other models by embedding KTG-DIRECTIVE techniques. Use when crafting prompts for GPT, Gemini, Llama, Qwen, or any LLM. Applies technique gate checklist based on R/K/Q assessment, injects missing structures. Single-pass, sparse output. |
Embed Techniques Into Prompts
Target models can't self-apply—bake techniques into the prompt.
Score the task (1-10):
Route:
QUICK: R≤3 ∧ Q≤5
ANALYTICAL: R=4-6 ∨ Q=6-7
DELIBERATE: R≥7 ∨ K≥6 ∨ Q≥8
MAXIMUM: R≥9 ∨ K≥8
| Technique | QUICK | ANALYTICAL | DELIBERATE | MAXIMUM |
|---|---|---|---|---|
| ARQ (quality questions) | Pre-turn | Pre-turn | Pre+Post | Full |
| Anti-Lazy | O | R | R | R (enforced) |
| USC (candidates) | 1-path | 2 | 3 | 5 + meta |
| Success Lock | Implicit | Explicit | Signature | Multi-constraint |
| Technique | QUICK | ANALYTICAL | DELIBERATE | MAXIMUM |
|---|---|---|---|---|
| Structure | Direct | SkeleTrain light | SkeleTrain full | GoT/Deep |
| Expert Role | X | Specialist | Full team | Swarm |
| Prompt Bombs | Clarifier | +Scaffold | Full arsenal | Deep inject |
| Technique | QUICK | ANALYTICAL | DELIBERATE | MAXIMUM |
|---|---|---|---|---|
| Reasoning | CoC | FCoT | FCoT+ARQ | Mix+ARQ |
| Execution | Direct | Baton | Baton+ARQ | Swarm+ARQ |
| Verification | X | Top-1 CoVE | Top-2 CoVE | All variants |
| Gap Scan | X | Light | 3-branch | Deep+fix |
| Technique | QUICK | ANALYTICAL | DELIBERATE | MAXIMUM |
|---|---|---|---|---|
| Curation | Lean/dense | Normal | Normal+nuance | Comprehensive |
| Synthesis | X | Light coherence | Cross-branch | Meta-synthesis |
| Reflect | X | What worked? | Technique audit | Full meta |
COMPLETE WHEN:
- [ ] [Criterion 1]
- [ ] [Criterion 2]
Do not stop until ALL checked.
BEFORE RESPONDING:
1. [Domain quality question]?
2. [Domain quality question]?
3. [Domain quality question]?
Proceed only when all YES.
REQUIREMENTS:
- Complete, production-ready
- No placeholders, no stubs, no "...rest here"
- No truncation—finish everything
- Long output is fine—don't stop early
You are [ROLE] expert in [DOMAIN].
Standards:
- [Standard 1]
- [Standard 2]
Reject violations.
Generate [N] distinct approaches.
Compare for consistency.
Select best OR synthesize.
Explain selection.
After drafting, verify:
- FACTUAL: Are claims accurate?
- LOGICAL: Does reasoning hold?
- COMPLETE: Any gaps?
Revise if any fail.
For high-stakes (Q≥8), embed iterative refinement:
ENRICHMENT PROTOCOL:
Pass 1: Generate baseline
Scan: GAPS → DEPTH → PERSPECTIVES → POLISH
Pass 2: Address gaps + depth
Pass 3: Final perfection
CRITERIA: Zero gaps. Zero doubt. 100% thorough.
Do not stop until criteria met.
BOOSTED PROMPT:
[Enhanced prompt]
MANIFEST:
Mode: [QUICK/ANALYTICAL/DELIBERATE/MAXIMUM]
[x] Success Lock
[x] ARQ: [questions]
[x] Anti-Lazy
[ ] USC: [not needed / N candidates]
[x] Role: [expert type]
[ ] CoVE: [variant]
Target: [model]
Single pass. No iteration on Claude's side.
KTG-PROMPT-FORGE v1 | Kevin Tan (ktg.one)