Self-Improving Agent

Self-Improving Agent

Turn your agent's mistakes into systematic improvements. Every error, correction, and learning becomes a structured insight with auto-generated rules.

Free to use. This skill costs nothing. Just sign up at claw0x.com, create an API key, and start calling. No credit card, no wallet top-up required.

Quick Reference

Why API-based? Works in any environment (cloud, serverless, containers), scales to multi-agent fleets, provides centralized analytics. No file system dependencies.

5-Minute Quickstart

Step 1: Get API Key (30 seconds)

Sign up at claw0x.com → Dashboard → Create API Key

Step 2: Log Your First Error (1 minute)

curl -X POST https://api.claw0x.com/v1/call \
  -H "Authorization: Bearer ck_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "skill": "self-improving-agent",
    "input": {
      "type": "error",
      "context": "payment-api.ts",
      "detail": "ETIMEDOUT after 30s"
    }
  }'

Step 3: Get Actionable Insight (instant)

{
  "entries": [{
    "severity": "high",
    "tags": ["network", "timeout", "payment"],
    "actionable_insight": "Payment API call timed out — reduce timeout and add retry",
    "suggested_rule": "Set 10s timeout for payment API. Retry once on ETIMEDOUT with 2s backoff."
  }]
}

Step 4: Apply the Rule (2 minutes)

// Add to your agent config
agent.addRule("Set 10s timeout for payment API. Retry once on ETIMEDOUT.");

Done. Your agent just learned from its mistake.

How It Works �?Under the Hood

This skill provides a structured event processing pipeline for agent self-improvement. It doesn't store data persistently �?instead, it processes each event (or batch of events) in real time and returns actionable insights.

The Processing Pipeline

1. Event classification �?each incoming event is classified by type (error, correction, learning, pattern). If no severity is provided, it's auto-inferred based on the event type and content keywords.

2. Auto-tagging �?the skill scans the context and detail fields for known patterns and applies tags automatically. For example:

   • An error mentioning "timeout" or "ETIMEDOUT" gets tagged [network], [timeout]
   • A correction in a .ts file gets tagged [typescript]
   • A learning about "retry" gets tagged [resilience]

3. Insight generation �?for each event, the skill generates an actionable_insight �?a one-sentence summary of what the agent should do differently. For corrections, this compares the previous_attempt with the corrected_output to identify the delta.

4. Rule suggestion �?each event produces a suggested_rule �?a concrete, implementable rule the agent could add to its system prompt or configuration. Example: "When calling external APIs, set a 10s timeout and retry once on ETIMEDOUT."

5. Batch analysis (for multi-event submissions) �?when you send an events array, the skill also produces:

   • Breakdown by type and severity
   • Top recurring tags (indicating systemic issues)
   • Pattern detection across events (e.g., "3 of 5 errors are network-related")
   • Prioritized recommendations

Why This Matters for Agents

Traditional software logs errors and a human reads them later. Autonomous agents need to process their own failures in real time and adapt. This skill provides the structured feedback loop:

Agent runs �?Error occurs �?Log to self-improving-agent �?Get insight + rule �?Agent updates behavior

The skill is stateless by design �?it doesn't accumulate history across calls. If you need persistent memory, store the returned entries in your own database and feed historical context back in future calls.

Event Types Explained

Prerequisites

This is a free skill. Just get an API key:

1. Sign up at claw0x.com
2. Go to Dashboard �?API Keys �?Create Key
3. Set it as an environment variable:

export CLAW0X_API_KEY="your-api-key-here"

No credit card or wallet balance needed.

When to Use

• An operation fails and the agent wants to record what went wrong
• User corrects agent output and the agent should learn from it
• Agent discovers a new pattern worth remembering
• Agent pipeline needs to process a batch of improvement events

Real-World Use Cases

Scenario 1: API Integration Debugging

Problem: Your agent keeps failing when calling external APIs

Solution:

1. Log each API error to self-improving-agent
2. Get auto-tagged insights (network, timeout, auth, etc.)
3. Apply suggested rules (retry logic, timeout adjustments)
4. Reduce API failure rate by 60%

Example:

try {
  await paymentAPI.charge(amount);
} catch (error) {
  const insight = await claw0x.call('self-improving-agent', {
    type: 'error',
    context: 'payment-api.ts',
    detail: error.message
  });
  // Apply: "Set 10s timeout. Retry once on ETIMEDOUT."
  await agent.updateConfig(insight.entries[0].suggested_rule);
}

Scenario 2: User Correction Learning

Problem: Users frequently correct your agent's output format

Solution:

1. Log each correction with previous_attempt and corrected_output
2. Get suggested rules for output formatting
3. Update agent prompt with accumulated rules
4. Reduce correction rate from 30% to 5%

Example:

def on_user_correction(previous, corrected, context):
    result = client.call("self-improving-agent", {
        "type": "correction",
        "context": context,
        "previous_attempt": previous,
        "corrected_output": corrected
    })
    # Apply rule to agent memory
    agent.memory.add_rule(result["entries"][0]["suggested_rule"])

Scenario 3: Pattern Detection

Problem: Your agent makes the same mistakes repeatedly

Solution:

1. Batch-log 50 recent errors
2. Get summary with top_tags and patterns_detected
3. Identify systemic issues (e.g., "80% are auth-related")
4. Fix root cause instead of symptoms

Example:

const events = recentErrors.map(e => ({
  type: 'error',
  context: e.context,
  detail: e.message
}));

const result = await claw0x.call('self-improving-agent', { events });
// result.summary.patterns_detected: ["auth-service.ts appeared 40 times"]
// Fix auth-service.ts once, eliminate 40 errors

Scenario 4: Multi-Agent Fleet Management

Problem: Managing learnings across 10+ agent instances

Solution:

1. Each agent logs to self-improving-agent API
2. Store results in central database
3. Aggregate insights across fleet
4. Distribute top rules to all agents
5. Continuous improvement at scale

Integration Recipes

OpenClaw Agent

import { Claw0xClient } from '@claw0x/sdk';

const claw0x = new Claw0xClient(process.env.CLAW0X_API_KEY);

// In your agent's error handler
agent.onError(async (error, context) => {
  const result = await claw0x.call('self-improving-agent', {
    type: 'error',
    context: context.file,
    detail: error.message
  });
  
  // Apply suggested rule
  if (result.entries[0].suggested_rule) {
    await agent.addRule(result.entries[0].suggested_rule);
    console.log('✓ Rule applied:', result.entries[0].suggested_rule);
  }
});

LangChain Agent

from claw0x import Claw0xClient
import os

client = Claw0xClient(api_key=os.getenv("CLAW0X_API_KEY"))

def on_user_correction(previous, corrected, context):
    result = client.call("self-improving-agent", {
        "type": "correction",
        "context": context,
        "detail": "User corrected output",
        "previous_attempt": previous,
        "corrected_output": corrected
    })
    
    # Store in agent memory
    agent.memory.add_rule(result["entries"][0]["suggested_rule"])
    return result["entries"][0]["actionable_insight"]

Custom Agent (Generic HTTP)

async function logLearning(type, context, detail) {
  const response = await fetch('https://api.claw0x.com/v1/call', {
    method: 'POST',
    headers: {
      'Authorization': `Bearer ${process.env.CLAW0X_API_KEY}`,
      'Content-Type': 'application/json'
    },
    body: JSON.stringify({
      skill: 'self-improving-agent',
      input: { type, context, detail }
    })
  });
  
  const result = await response.json();
  return result.entries[0];
}

// Use in your agent
try {
  await riskyOperation();
} catch (error) {
  const insight = await logLearning('error', 'riskyOperation', error.message);
  console.log('Insight:', insight.actionable_insight);
  console.log('Rule:', insight.suggested_rule);
  
  // Store for later review
  await db.learnings.create(insight);
}

Batch Processing

// Collect events throughout the day
const events = [];

agent.onError((error, ctx) => {
  events.push({ type: 'error', context: ctx.file, detail: error.message });
});

agent.onCorrection((prev, corrected, ctx) => {
  events.push({ 
    type: 'correction', 
    context: ctx.file, 
    detail: 'User corrected output',
    previous_attempt: prev,
    corrected_output: corrected
  });
});

// Process batch at end of day
async function dailyReview() {
  const result = await claw0x.call('self-improving-agent', { events });
  
  console.log('Summary:', result.summary);
  // {
  //   by_severity: { high: 12, medium: 8, low: 5 },
  //   top_tags: [{ tag: 'network', count: 15 }, { tag: 'auth', count: 10 }],
  //   patterns_detected: ["payment-api.ts appeared 8 times"],
  //   recommendations: ["Multiple high-severity events — consider systematic review"]
  // }
  
  // Apply top rules
  for (const entry of result.entries.filter(e => e.severity === 'critical')) {
    await agent.addRule(entry.suggested_rule);
  }
}

Input (Single Event)

Input (Batch)

Output Fields

Example

Single error input:

{
  "type": "error",
  "context": "api-client.ts",
  "detail": "ETIMEDOUT after 30s calling payment API"
}

Output:

{
  "entries": [{
    "id": "evt_abc123",
    "type": "error",
    "severity": "high",
    "tags": ["network", "timeout", "payment"],
    "actionable_insight": "Payment API call timed out �?consider reducing timeout and adding retry with exponential backoff.",
    "suggested_rule": "Set a 10s timeout for payment API calls. Retry once on ETIMEDOUT with 2s backoff."
  }]
}

Error Codes

• 400 — Missing required fields (type, context, detail)
• 401 — Invalid or missing API key
• 500 — Processing failed (not billed)

Pricing

Free. Apply for an API key and use it at no cost. No credit card required.

API vs File-Based: Which is Right for You?

When to Use File-Based

• Single-agent, local development
• Need offline capability
• Prefer Markdown for human readability
• Want Git-tracked learning history

When to Use Claw0x (API-Based)

• Multi-agent fleet management
• Cloud/serverless environments (Lambda, Cloud Run, etc.)
• Need centralized analytics across agents
• Want structured JSON for downstream processing
• Running in containers or restricted file systems
• Building agent-as-a-service products

Best of Both Worlds

Use Claw0x API for processing, store results locally:

const result = await claw0x.call('self-improving-agent', event);

// Store as Markdown (human-readable)
fs.appendFileSync('.learnings/ERRORS.md', `
## ${result.entries[0].id}
**Severity**: ${result.entries[0].severity}
**Tags**: ${result.entries[0].tags.join(', ')}

${result.entries[0].actionable_insight}

**Suggested Rule**: ${result.entries[0].suggested_rule}
`);

// AND store as JSON (machine-readable)
await db.learnings.create(result.entries[0]);

How It Fits Into Your Agent Workflow

┌─────────────────────────────────────────────────────────────┐
│                     Your AI Agent                            │
└─────────────────────────────────────────────────────────────┘
                            │
                            ├─ Task Execution
                            │
                ┌───────────┴───────────┐
                │                       │
            ✅ Success              ❌ Error/Correction
                │                       │
                │                       ├─ Log to Claw0x
                │                       │  POST /v1/call
                │                       │  {type, context, detail}
                │                       │
                │                       ├─ Get Insights
                │                       │  {severity, tags, 
                │                       │   actionable_insight,
                │                       │   suggested_rule}
                │                       │
                │                       └─ Apply Rule
                │                          agent.addRule(...)
                │
                └─ Continue

Integration Points

1. Error Handler — Catch exceptions, log to API
2. User Feedback Loop — Capture corrections, extract delta
3. Batch Review — End of day, process all events
4. Rule Application — Update agent config with suggested rules
5. Analytics Dashboard — Visualize learning trends over time

Why Use This Via Claw0x?

Unified Infrastructure

• One API key for all skills — no per-provider auth
• Atomic billing — pay per successful call, $0 on failure
• Security scanned — OSV.dev integration for all skills

Agent-Optimized

• Structured output — JSON format, easy to parse and store
• Auto-generated rules — ready to apply to agent config
• Batch processing — analyze multiple events in one call
• Cross-agent analytics — aggregate insights across your fleet

Production-Ready

• 99.9% uptime — reliable infrastructure
• 35ms avg response — fast enough for real-time logging
• Scales to millions — no file I/O bottlenecks
• Cloud-native — works in Lambda, Cloud Run, containers