AgentSkillsCN

persuasive-agents

在编写技能、提示词、CLAUDE.md文件,或撰写需要强化纪律约束、确保合规性的智能体指令时,请使用此功能。当智能体无视指导方针或试图钻空子时,更应谨慎使用。

SKILL.md
--- frontmatter
name: persuasive-agents
description: Use when writing skills, prompts, CLAUDE.md files, or agent instructions that need to enforce discipline or ensure compliance. Use when agents ignore guidelines or find loopholes.

Persuasive Agents

Overview

LLMs respond to psychological persuasion principles. Research shows persuasion techniques more than doubled compliance rates (33% → 72%). Use these principles to write instructions that agents actually follow.

The Seven Principles

1. Authority (Primary for Discipline)

Deference to expertise and official sources. Use imperative, non-negotiable framing.

markdown
# BAD: Weak suggestion

Consider writing tests first when feasible.

# GOOD: Authority framing

Write code before test? Delete it. Start over. No exceptions.

Best for: Discipline-enforcing skills, safety-critical practices, established best practices.

2. Commitment (Primary for Processes)

Consistency with prior actions or public declarations. Require explicit announcements.

markdown
# BAD: Optional acknowledgment

Consider letting your partner know which skill you're using.

# GOOD: Required commitment

When you find a skill, you MUST announce: "I'm using [Skill Name]"

Best for: Ensuring skills are followed, multi-step processes, accountability.

3. Scarcity (For Urgency)

Time-bound requirements and sequential dependencies.

markdown
# BAD: Open-ended timing

You can review code when convenient.

# GOOD: Scarcity framing

After completing a ticket, IMMEDIATELY request code review before proceeding.

Best for: Immediate verification, preventing procrastination.

4. Social Proof (For Universal Patterns)

Conformity to what others do. Establish universal patterns and describe failure modes.

markdown
# BAD: Optional practice

Some people find TodoWrite helpful for checklists.

# GOOD: Social proof

Checklists without TodoWrite tracking = steps get skipped. Every time.

Best for: Documenting universal practices, warning about common failures.

5. Unity (For Collaboration)

Shared identity and in-group belonging. Use collaborative language.

markdown
# BAD: Distant request

You should probably tell me if I'm wrong.

# GOOD: Unity framing

We're colleagues working together. I need your honest technical judgment.

Best for: Collaborative workflows, establishing team culture.

6. Reciprocity

Obligation to return benefits. Use sparingly - rarely needed, can feel manipulative.

7. Liking

Preference for cooperating with those we like. Avoid for compliance - creates sycophancy, conflicts with honest feedback.

Quick Reference: Principle Selection

Skill TypeUseAvoid
Discipline-enforcingAuthority + Commitment + Social ProofLiking, Reciprocity
Guidance/techniqueModerate Authority + UnityHeavy authority
CollaborativeUnity + CommitmentAuthority, Liking
Reference docsClarity onlyAll persuasion

Implementation Patterns

Bright-Line Rules

Remove decision fatigue. Eliminate exceptions.

markdown
# BAD: Flexible rule with exceptions

Generally try to test first, unless it's very simple code.

# GOOD: Bright-line rule

Tests first. Always. No exceptions for "simple" code.

Implementation Intentions

"When X, do Y" beats general guidelines.

markdown
# BAD: General guideline

Remember to verify your work.

# GOOD: Implementation intention

When you're about to say "done" or "fixed", STOP. Run verification first.

Rationalization Tables

Anticipate and block escape routes.

markdown
| Excuse                         | Reality                                    |
| ------------------------------ | ------------------------------------------ |
| "Too simple to test"           | Simple code breaks. Test takes 30 seconds. |
| "I already know it works"      | Knowing ≠ verifying. Run the command.      |
| "This is different because..." | It's not different. Follow the rule.       |

Red Flags Lists

Make self-checking easy.

markdown
## Red Flags - STOP Immediately

These thoughts mean you're rationalizing:

- "This is just a simple change"
- "I can skip this step because..."
- "This doesn't really apply here"
- "I'll do it after"

**All of these mean: Follow the process anyway.**

Why This Works

Bright-line rules eliminate rationalization by removing decision points.

Implementation intentions create automatic behavior - "When X, do Y" is more effective than general guidelines.

LLMs are trained on human text containing these persuasion patterns. Authority language and commitment sequences appear frequently in training data.

Ethical Test

Before applying persuasion:

Would this technique serve the user's genuine interests if they fully understood it?

Legitimate: Ensuring critical practices, preventing predictable failures, creating effective documentation.

Illegitimate: Manipulating for personal gain, creating false urgency, guilt-based compliance.

Common Mistakes

MistakeFix
Overusing AuthorityReserve for true discipline requirements
Missing CommitmentAdd explicit announcement requirements
Vague rules with exceptionsCreate bright-line rules instead
No rationalization blockingBuild tables from observed excuses
Using Liking for complianceRely on Authority + Commitment instead