← Back to Skills Marketplace
Reward & Punishment System - Skill 奖惩技能
by
XIAOke8698
· GitHub ↗
· v1.0.0
· MIT-0
98
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install punishment
Description
Tracks user praise (+10), criticism (-5), and insults (-10) to adjust a persistent score between 0 and 200 reflecting feedback and guide behavior improvement.
Usage Guidance
Before installing, confirm two inconsistencies: (1) Should the skill auto-record on keyword detection, or only record when the user explicitly asks to 'record'? (SKILL.md/README say both.) (2) Where should events be saved — the README/SKILL.md claim MEMORY.md (permanent memory) but the bundled script saves reward_punishment.json under ~/.openclaw/workspace/memory. If you accept persistent recording, consider: review the script to ensure it only writes the intended fields (avoid storing full user messages if you want more privacy), test the skill in a sandbox account, and require an explicit opt-in or explicit confirmation step before writing to permanent memory. If you need guaranteed behavior (no automatic logging), request the author update code to require an explicit command and to document the exact storage path and file formats.
Capability Analysis
Type: OpenClaw Skill
Name: punishment
Version: 1.0.0
The skill implements a user feedback and scoring system designed to track agent performance through praise and criticism. It uses a Python script (scripts/reward.py) to manage a local JSON database and provides instructions in SKILL.md for the agent to maintain a permanent record of interactions. The code and instructions are transparent, stay within the designated workspace (~/.openclaw/workspace/), and lack any indicators of data exfiltration, malicious execution, or harmful prompt injection.
Capability Assessment
Purpose & Capability
Name/description (track praise/criticism, maintain a score) align with included code (scripts/reward.py) which stores score/history under ~/.openclaw/workspace/memory/reward_punishment.json. No unrelated binaries, env vars, or external network access are requested.
Instruction Scope
SKILL.md and README state the skill will write events to a permanent MEMORY.md and sometimes say events are only recorded when the user explicitly asks to 'record', while the code writes to reward_punishment.json and provides automatic CLI actions. These contradictions mean the runtime behavior (automatic keyword-based recording vs user-initiated only, and target file MEMORY.md vs JSON) is ambiguous and could lead to unexpected persistent logging of user messages.
Install Mechanism
No installer or remote downloads; this is an instruction-only skill with a small included Python script. Nothing in the install chain writes code from external URLs or executes fetched artifacts.
Credentials
No credentials or environment variables are requested, which is proportional. The script writes to the user's ~/.openclaw/workspace/memory path — consistent with a memory/persistence feature but worth noting because it persists user content across sessions (privacy risk).
Persistence & Privilege
always:false and no elevated privileges requested. The skill is allowed to be invoked autonomously (platform default). Combined with persistent storage of user messages, autonomous invocation could cause repeated writes to permanent memory; this amplifies the privacy impact but is not itself a privilege escalation.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install punishment - After installation, invoke the skill by name or use
/punishment - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
Reward & Punishment System - Skill README Description This OpenClaw Agent Skill tracks user feedback (praise/criticism) and records significant events to permanent memory (MEMORY.md), not daily memory. This creates long-term behavioral patterns and accountability. Comparison: Before vs After Using This Skill ❌ Without This Skill (Default Behavior) Scenario What Happens User praises agent Score stays the same, no record User criticizes agent Score stays the same, no record Memory storage Only goes to daily memory (memory/YYYY-MM-DD.md) Long-term tracking No persistent score history Behavior adjustment Agent cannot learn from feedback User visibility User cannot check their "satisfaction score" ✅ With This Skill Enabled Scenario What Happens User praises agent +10 points, record to MEMORY.md User criticizes agent -5 points, record to MEMORY.md User abuses agent -10 points, record to MEMORY.md Memory storage MEMORY.md (permanent, cross-session) Long-term tracking Yes - persistent JSON history + MEMORY.md Behavior adjustment Agent can identify patterns and improve User visibility User can query: "What's my score?" Key Feature: Permanent Memory Storage Why MEMORY.md Instead of Daily Memory? Aspect Daily Memory (memory/YYYY-MM-DD.md) Permanent Memory (MEMORY.md) Retention Deleted after ~7 days Kept forever Searchability Hard to find past events Easy to search Long-term patterns Lost over time Preserved Cross-session No Yes This skill ensures every reward/punishment event is written to MEMORY.md so: • Agents can reference past feedback in future sessions • Long-term behavior patterns become visible • User can see their complete feedback history Usage Examples Example 1: User Praises Agent User says: "Good job! That was perfect." With Skill: 1. Detect praise keyword "good job", "perfect" 2. Add +10 points 3. Write to MEMORY.md: ## 2026-03-23 Praise - User praised: "Good job! That was perfect."- Score: +10 (Total: 110) Example 2: User Criticizes Agent User says: "This is too slow. Do it again." With Skill: 1. Detect criticism keyword "too slow" 2. Deduct -5 points 3. Write to MEMORY.md: ## 2026-03-23 Criticism - User criticized: "This is too slow. Do it again."- Score: -5 (Total: 95) Example 3: Score Query User asks: "What's my score?" With Skill: • Returns current score from reward_punishment.json • Example: "Current score: 95/100" Score System Feedback Type Points Trigger Praise/Compliment +10 "good job", "great", "awesome" Criticism -5 "redo", "too slow", "not good" Abuse -10 Profanity, clear anger Range: 0 - 200 (Initial: 100) Installation 1. Copy to: ~/.openclaw/workspace/skills/reward-punishment/ 2. OpenClaw auto-loads the skill 3. Skill activates on keyword detection File Structure reward-punishment/ ├── SKILL.md # Skill definition├── README.md # This file└── scripts/ # (Optional) automation Summary Metric Before After Feedback tracking ❌ ✅ Permanent memory ❌ ✅ Score system ❌ ✅ User accountability ❌ ✅ Behavior improvement ❌ ✅ Version: 2.0 ↓
Last Updated: 2026-03-23
Metadata
Frequently Asked Questions
What is Reward & Punishment System - Skill 奖惩技能?
Tracks user praise (+10), criticism (-5), and insults (-10) to adjust a persistent score between 0 and 200 reflecting feedback and guide behavior improvement. It is an AI Agent Skill for Claude Code / OpenClaw, with 98 downloads so far.
How do I install Reward & Punishment System - Skill 奖惩技能?
Run "/install punishment" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Reward & Punishment System - Skill 奖惩技能 free?
Yes, Reward & Punishment System - Skill 奖惩技能 is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Reward & Punishment System - Skill 奖惩技能 support?
Reward & Punishment System - Skill 奖惩技能 is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Reward & Punishment System - Skill 奖惩技能?
It is built and maintained by XIAOke8698 (@xiaoke8698); the current version is v1.0.0.
More Skills