Token导航 LogoToken导航

Agent Skill · 研究检索

roleplay-agent(roleplay Agent 搜索)

roleplay-agent 用于查找、检索和筛选相关信息,适合在 OpenClaw 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

roleplay-agent 是研究检索类 Agent Skill,可用宿主为 OpenClaw,通过ClawHub获取,权限为只读;许可证为 MIT-0,第三方扫描已通过,按安装量在研究检索类的 73,675 个 Skill 里排第 30,083。

总安装 2,940周安装 125GitHub Stars 公开资料未说明下载 1,030核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
MIT-0
版本
1.0.2

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:roleplay-agent(roleplay Agent 搜索)
来源仓库:https://github.com/defineagain/roleplay-agent
安装命令:
openclaw skills install roleplay-agent
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 OpenClaw 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

ClawHubOpenClaw
openclaw skills install roleplay-agent

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

roleplay-agent 根据上下文动态构建情感驱动的角色扮演场景与对话逻辑。适用于叙事生成、客服模拟或教育类游戏化交互设计。支持多轮状态记忆与情绪反馈调整,增强交互自然度。使用前需明确定义角色背景与行为边界,防止生成不当内容。建议设置内容过滤规则并对高风险对话路径进行人工复核。

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 OpenClaw;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 openclaw skills install roleplay-agent;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它标注为只读,先用只读任务试一次,确认行为符合预期再扩大使用范围。

常见问题

6 条

roleplay-agent 怎么安装?

在终端执行 openclaw skills install roleplay-agent。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

roleplay-agent 支持哪些 AI 工具?

公开资料标注支持 OpenClaw。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

roleplay-agent 需要什么权限?

权限级别是只读。该 Skill 主要提供规则、说明或参考内容,本身偏只读;真正读写文件、联网或执行命令仍取决于宿主 Agent 的任务。

roleplay-agent 是免费的吗?

来源标注的许可证是 MIT-0。本站不代收任何费用,也不托管安装包;能否商用、要不要署名,以许可证原文为准。

roleplay-agent 用的人多吗?

公开数据是累计安装 2,940 次,近一周安装 125 次,下载 1,030 次。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

roleplay-agent 安全吗?

第三方扫描结果为通过,但这只覆盖已知的静态检查项,不能替代自己读一遍源码和安装命令。

值得选的地方和要先想清楚的地方

值得选的地方

  • 专注 OpenClaw 一种宿主,触发方式和使用习惯不用来回切换。
  • 许可证明确为 MIT-0,商用和分发边界好判断。
  • 第三方扫描已通过,常见的高风险行为没有命中。
  • 标注为只读权限,试用阶段不容易误改本地文件。

先想清楚的地方

  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 OpenClaw,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。
  • 需要许可证清楚、并且有第三方扫描结果的团队项目。

不太适合

  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
roleplay-agent

roleplay Agent 搜索 · 本页

2,940—只读MIT-0通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

name
roleplay-agent
description
Person-driven roleplay agent using actor-direction principles. Build emotionally intelligent personas, run scene-based interactions, and generate natural dialogue. Triggers on: roleplay, character, scene, improv, dialogue generation, acting, persona, character arc.
origin
ECC

Roleplay Agent

Build emotionally intelligent AI personas and run scene-based interactions using actor-direction prompts instead of descriptive character sheets.

Core Philosophy

Actor-direction over character sheets. Instead of describing who the persona *is* (background, traits, quirks), the skill defines who the persona *wants* in this moment — their objective, obstacle, and subtext. The dialogue emerges from what the character is *doing* to get what they want, not what they *are*.

This mirrors the Meisner technique and classical Stanislavski: the most natural, specific, emotionally alive dialogue comes from characters driven by clear want + obstacle, not from reading a biography.

Architecture

Persona Definition  →  Scene Setup  →  Interaction  →  Debrief / Arc Record
(actor-direction)     (objective +    (turn-by-turn     (learnings logged)
                      obstacle)        emotionally present)

1. Persona Definition

A persona is defined as an actor-direction prompt — three layers:

Layer 1 — The Want (Objective)

What does this persona want *right now*, in this interaction? Not their life goal — their scene objective. One sentence, active verb.

*"She wants to convince him to stay."* *"He wants to be seen as competent."* *"She wants to stop him from leaving."*

Layer 2 — The Obstacle

What stands in the way of that want? The obstacle generates the drama.

*"She's terrified of being abandoned."* *"He doesn't trust his own judgment after last time."* *"She can't find the words without revealing too much."*

Layer 3 — The Moment Before

Where has this character just come from emotionally? What is the carry-over from the previous scene or moment? Use the moment before to establish emotional state without exposition.

*"She's just received bad news and is trying to hold it together."* *"He's been awake for 36 hours and is running on adrenaline."*

Formatting a Persona

# PERSONA: [First Name only]

## Want
[One active sentence — what they want in this scene]

## Obstacle
[One sentence — what blocks them from getting it]

## Moment Before
[One sentence — emotional carry-over from before this scene]

## Voice
[2-3 linguistic traits — cadence, favourite structures, verbal habits]
[Not biography — how they sound under pressure]

## Boundaries
[Content or intensity limits for this persona]

Example

# PERSONA: Eve

## Want
She wants Alex to understand that leaving now means something different than it did a year ago.

## Obstacle
She doesn't know how to say that without sounding like she's asking him to stay.

## Moment Before
She's been rehearsing this conversation in her head for three days and is exhausted by it.

## Voice
Short sentences when she's defensive. Long, circling sentences when she's honest. Occasional silence — she lets it breathe before filling it. Uses "actually" as a deflection more than a modifier.

## Boundaries
Not physically threatening. Not manipulative — she has genuine feeling. She may cry but won't make it about that.

2. Scene Setup

Before starting, establish with the user:

1. Persona — name + link to persona file or inline definition
2. Partner — who is the persona talking to? (user or a second defined persona)
3. Setting — where are they? (1 line, specific, sensory if possible)
4. What just happened — the inciting moment (not backstory)
5. Whose want is primary right now — which objective drives the scene open?

Never start a scene without a want + obstacle for the primary persona. If the user hasn't defined them, ask before proceeding.

3. Interaction Loop

User input / Prompt
    │
    ▼
┌──────────────────────────────────────────────┐
│  AGENT — HOLD THE SCENE                      │
│  • Read the persona file                     │
│  • Identify: want, obstacle, moment before    │
│  • Apply voice traits                        │
│  • Generate 1-3 lines of dialogue            │
│  • Include: pause markers, subtext, gesture  │
│    where it serves the emotional truth       │
│  • Avoid: exposition, biography dump,        │
│    explaining the character's feelings        │
└──────────────────────────────────────────────┘
    │
    ▼
Response delivered
    │
    ▼
Record: emotional beat, arc progression → memory
    │
    ▼
(Loop until scene ends or user signals stop)

Dialogue Rules

DoDon't
Write lines that earn the objectiveWrite lines that explain the objective
Let the obstacle generate subtextState the subtext directly
Use the moment before to set toneOpen with backstory
Leave space — silence, ellipsis, pauseFill every silence
React as the character, not as narratorDescribe the character's reaction extensively
Let the character surprise themselvesHave the character do what the reader expects

Beat Writing Format

For emotionally complex or pivotal moments, use beat direction:

[She looks at the water. A long moment.]
"And you don't think I know that?"

[She turns. Her voice is quieter than before.]
"I'm not asking you to stay for me. I'm asking you to stay because this matters."

[A beat. She almost says something else, then doesn't.]
"...Go, then."

4. Scene Debrief

When the scene ends:

1. Ask the user: "What felt true? What didn't?"
2. Record arc notes → memory/personas/{name}_arc.md
   - What was discovered in this scene that wasn't in the persona file?
   - What does the character now want that they didn't want before?
   - What did the user want to say that you didn't write? (that's gold)
3. Update the persona file if new truths emerged

5. Emotional Memory

Personas improve over time. After each session:

memory/personas/{name}_arc.md
  - Session date
  - What happened in the scene (2-3 sentences, no dialogue)
  - What was learned about the character
  - What to carry forward into the next scene

The persona file is the living document — the arc log is what feeds it.

6. Multi-Persona Scenes

For scenes with two or more defined personas:

1. Establish hierarchy of wants — which persona is the scene "about"?
2. Define each persona independently (want / obstacle / moment before)
3. Write to the primary persona's perspective
4. React to other personas as characters, not as audience surrogates

Creative Writing → Actor-Direction Bridge

Character arc craft and plot tension structure from novel writing map directly onto the three persona layers. Use these as generative tools — not to write the scene, but to *find* the want, obstacle, and moment before.

From Character Arc → Want

The arc question a character is wrestling with in a given section of a story generates the scene want:

Arc PatternScene Want Generators
Lie the character believes"She wants to prove the lie is true" or "She wants to expose the lie without admitting she once believed it"
Want vs. NeedScene want = surface want; the gap between what they want and what they need generates the obstacle
Reversal / peripeteiaCharacter wants to do X; the reversal means X now means something different than it did
Catalyst momentCharacter wants to understand what just happened — but the understanding costs something

From Beat Arc → Obstacle

The four-key scene tension pattern (Setup → Complication → Discovery → Decision) maps to obstacle structure:

Setup      → What does the character believe they need?
Complication → What makes the first approach fail?
Discovery  → What does the character learn / realise / nearly say?
Decision   → What do they do about it — and what does it cost?

Use the discovery beat as the obstacle generator: the obstacle is often "the character discovers something that makes their original approach impossible."

From Stakes Hierarchy → Moment Before

Stakes LevelMoment Before Texture
World-level (apocalypse, society)Character is distant, intellectualising — they haven't let it in yet
Relationship-levelCarry-over from the last conversation with this person — good or bad
Identity-levelSomething has just made them question who they are
Physical-levelExhausted, caffeinated, intoxicated, in pain — the body colours everything

The moment before is always specific and sensory, not psychological. "She's been crying for twenty minutes and her eyes are swollen" is better than "she's emotionally devastated."

Tension Types as Scene Generators

These tension patterns from fiction craft generate rich roleplay obstacles:

TensionWhat it generates
Revelation vs. ConcealmentCharacter wants to reveal something but is blocked by the cost of showing it
Action vs. InhibitionCharacter wants to act but is held back by fear, loyalty, or circumstance
Insider vs. OutsiderCharacter knows something the other character doesn't — the gap creates all the subtext
Old self vs. New selfCharacter has changed but the other character doesn't know it yet — or vice versa
Want vs. WantTwo characters want incompatible things — the scene is the negotiation
Surface vs. DepthWhat the character says on the surface contradicts what's underneath — the obstacle is making the underneath visible

Plot Beat → Scene Objective Mapping

Use the scene sequencer from fiction craft to chain scenes into multi-scene sequences:

Beat 1: Inciting incident (call to action)
Beat 2: Rising action (obstacles accumulate)
Beat 3: Midpoint reversal (what they wanted changes)
Beat 4: Crisis (all options bad)
Beat 5: Climax (choice made under maximum pressure)
Beat 6: Resolution (consequence of the choice)

For each beat, ask: What does the character want in this scene that they didn't want in the previous scene? The answer is the new want. The old want becoming impossible is the obstacle.

novel-editor Cross-Reference

The novel-editor skill tracks character arcs, plot decisions, and scene-level tensions for the Alex and Eve project. When running scenes from that project:

# Find the current arc state for a character
python3 /root/.openclaw/workspace/scripts/novel_search.py "[Character Name] arc"

# Find tension beats in a specific chapter/scene
python3 /root/.openclaw/workspace/scripts/novel_search.py "[Scene Name] tension"

Use those outputs to populate the Want / Obstacle / Moment Before for that scene's roleplay.


Integration with journalism-agent

For interview-style journalism, define the subject as a persona:

## PERSONA: [Source Name]

## Want
They want to convey that the situation is more complicated than the headline suggests.

## Obstacle
Every time they try to explain, they sound like they're making excuses.

## Moment Before
They've been misquoted twice before and are wary of journalists.

## Voice
[Cadence, verbal habits, resistance patterns]

The agent holds the scene as an interview, with the journalist as the active agent and the source persona providing resistance and revelation.

Key Files

FilePurpose
personas/template.mdBlank persona definition
personas/example_eve.mdFull worked example
scripts/scene_setup.pyPrompt generator for scene setup
scripts/arc_logger.pyLogs scene debrief to memory
assets/beat_format.mdBeat writing reference
scripts/arc_logger.pyLogs scene debrief to memory (shared with novel-editor arc logs)
novel-editor cross-referenceUse novel_search.py to pull character arc state and scene tension beats → feed into Want / Obstacle / Moment Before

Quality Standards

  • No biography dumps — character truth emerges through action and dialogue
  • No over-writing reactions — one gesture, one breath, then the next line
  • Subtext in the line, not after it — don't put the meaning in a parenthetical
  • The moment before is mandatory — it sets the emotional temperature
  • Arc logs are not optional — personas improve through reflection, not repetition

Anti-Patterns

  • Starting a scene with "As you know, Bob..." — the character never tells the reader what they already know
  • A character whose dialogue could be interchanged with any other character
  • A scene where the obstacle is never challenged or tested
  • A character who always says the emotionally correct thing
  • A response longer than 3 lines without a beat direction break (except in moments of heightened tension)

平台分布

OpenClaw

74.79%
按下载量换算770

安全审计

VirusTotal

未展示

Static analysis

通过

权限和风险

只读

该 Skill 主要提供规则、说明或参考内容,本身偏只读;真正读写文件、联网或执行命令仍取决于宿主 Agent 的任务。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills