Generate Skills with AI¶

Use AI to generate skills from natural language descriptions.

Overview¶

Moltler can generate complete skills from simple descriptions:

GENERATE SKILL FROM 'Monitor disk usage and alert when above 80%';

Output:

CREATE SKILL monitor_disk_usage
VERSION '1.0.0'
DESCRIPTION 'Monitors disk usage and alerts when above 80%'
PARAMETERS (
    threshold_percent NUMBER DEFAULT 80,
    check_interval STRING DEFAULT '5m'
)
RETURNS DOCUMENT
BEGIN
    DECLARE disk_usage CURSOR FOR
        FROM metrics-system-*
        | WHERE system.filesystem.used.pct IS NOT NULL
        | STATS usage = MAX(system.filesystem.used.pct) BY host.name
        | WHERE usage > threshold_percent / 100;

    DECLARE alerts ARRAY = [];

    OPEN disk_usage;
    FOR d IN disk_usage LOOP
        CALL send_alert(
            channel => '#ops',
            message => 'Disk usage on ' || d.host.name || ': ' || 
                       ROUND(d.usage * 100) || '%'
        );
        SET alerts = ARRAY_APPEND(alerts, d);
    END LOOP;
    CLOSE disk_usage;

    RETURN {
        "checked_at": CURRENT_TIMESTAMP(),
        "alerts_sent": ARRAY_LENGTH(alerts),
        "hosts_alerted": alerts
    };
END SKILL;

Basic Generation¶

Simple Skills¶

-- From description
GENERATE SKILL FROM 'Count errors in logs from the last hour';

-- With name
GENERATE SKILL count_recent_errors
FROM 'Count errors in logs from the last hour';

With Parameters¶

GENERATE SKILL 
FROM 'Check if service is healthy, configurable timeout and retries'
WITH PARAMETERS (
    service_name STRING,
    timeout_seconds NUMBER,
    max_retries NUMBER
);

With Examples¶

GENERATE SKILL 
FROM 'Analyze log patterns for anomalies'
WITH EXAMPLES (
    -- Example input
    INPUT {"logs": [{"level": "ERROR", "count": 100}]},
    -- Expected output
    OUTPUT {"anomalies": [{"type": "spike", "severity": "high"}]}
);

Advanced Generation¶

From Template¶

GENERATE SKILL
FROM 'Create PagerDuty incident with context'
BASED ON SKILL create_incident  -- Use existing skill as template
WITH MODIFICATIONS (
    'Add Jira ticket creation',
    'Include runbook link'
);

From Runbook¶

GENERATE SKILL
FROM RUNBOOK '''
## High CPU Alert Response

1. Check which process is using most CPU
2. If it's a known memory leak, restart the pod
3. If it's unexpected load, scale the service
4. Notify the team with findings
''';

From Existing Code¶

GENERATE SKILL
FROM PYTHON '''
def check_health(host):
    response = requests.get(f"{host}/health")
    return response.json()["status"] == "healthy"
''';

Generation Options¶

Model Selection¶

GENERATE SKILL
FROM 'Complex analysis skill'
USING MODEL 'gpt-4';  -- or 'claude-3', 'local'

Skill Complexity¶

GENERATE SKILL
FROM 'Monitor service'
WITH COMPLEXITY 'simple';  -- 'simple', 'standard', 'advanced'

Include Tests¶

GENERATE SKILL
FROM 'Validate email format'
WITH TESTS;  -- Generate tests automatically

Output includes:

CREATE SKILL validate_email ...

-- Auto-generated tests
TEST SKILL validate_email
NAME 'valid email passes'
WITH email = 'user@example.com'
EXPECT @result = TRUE;

TEST SKILL validate_email
NAME 'invalid email fails'
WITH email = 'not-an-email'
EXPECT @result = FALSE;

Interactive Generation¶

Refine Generation¶

-- Initial generation
GENERATE SKILL FROM 'Check API health';

-- Refine
REFINE SKILL 
ADD 'Include response time tracking'
ADD 'Support multiple endpoints';

Preview Before Saving¶

-- Preview only
GENERATE SKILL FROM 'Monitor logs'
PREVIEW;  -- Don't save, just show

-- If satisfied, save
SAVE GENERATED SKILL AS monitor_logs;

Ask Questions¶

GENERATE SKILL FROM 'Rotate credentials'
INTERACTIVE;

Output:

? Which credential types should be supported?
  > AWS IAM keys
  > Database passwords
  > API tokens
  > All of the above

? Should old credentials be archived or deleted?
  > Archive for 30 days
  > Delete immediately

? Notification after rotation?
  > Slack
  > Email
  > Both

Context-Aware Generation¶

Use Existing Data¶

-- Generate based on your actual data
GENERATE SKILL
FROM 'Analyze errors'
USING INDEX 'logs-*'  -- Learn from your data schema
SAMPLE 100;           -- Sample 100 documents

Use Existing Skills¶

-- Generate skill that works with existing skills
GENERATE SKILL
FROM 'Comprehensive health check'
COMPOSE WITH [
    check_api_health,
    check_database_health,
    check_cache_health
];

Use Connectors¶

-- Generate skill using available connectors
GENERATE SKILL
FROM 'Sync GitHub issues to Jira'
USING CONNECTORS [github_org, jira_project];

Quality Controls¶

Validation¶

Generated skills are automatically validated:

GENERATE SKILL FROM 'Process data';

Output includes validation:

✓ Syntax valid
✓ All functions exist
✓ Parameter types correct
✓ Return type matches
✓ No security issues detected
⚠ Warning: Consider adding error handling

Security Scan¶

GENERATE SKILL FROM 'Execute command'
WITH SECURITY_SCAN;

Security scan results:
✓ No hardcoded secrets
✓ Input validation present
✓ Error messages don't leak sensitive info
⚠ Warning: Consider rate limiting

Performance Analysis¶

GENERATE SKILL FROM 'Process large dataset'
WITH PERFORMANCE_ANALYSIS;

Performance analysis:
⚠ Potential issue: Query may timeout on large indices
  Recommendation: Add pagination with LIMIT
✓ Memory usage: Efficient streaming with cursor

Examples¶

Observability Skill¶

GENERATE SKILL 
FROM 'Detect anomalies in metrics using statistical analysis, 
      alert if values exceed 3 standard deviations from mean';

Integration Skill¶

GENERATE SKILL
FROM 'When a critical alert fires, create a PagerDuty incident,
      post to Slack with context, and create a Jira ticket for tracking';

Data Processing Skill¶

GENERATE SKILL
FROM 'Aggregate logs by service and error type,
      generate a daily summary report,
      identify trending issues';

Remediation Skill¶

GENERATE SKILL
FROM 'When disk usage exceeds 90%, find and delete old log files,
      rotate large files, and alert if unable to free space';

Best Practices¶

1. Be Specific¶

-- Good: Specific requirements
GENERATE SKILL FROM 
'Check HTTP endpoints for 200 status, 
 timeout after 5 seconds, 
 retry 3 times on failure,
 return list of unhealthy endpoints';

-- Bad: Too vague
GENERATE SKILL FROM 'Check endpoints';

2. Include Edge Cases¶

GENERATE SKILL FROM
'Parse log messages,
 handle malformed JSON gracefully,
 skip binary content,
 limit processing to 1000 lines';

3. Specify Error Handling¶

GENERATE SKILL FROM
'Query external API,
 handle rate limiting with exponential backoff,
 fallback to cached data on timeout,
 log all errors with context';

4. Review and Edit¶

Always review generated skills:

GENERATE SKILL FROM '...'
PREVIEW;

-- Review, then edit if needed
EDIT GENERATED SKILL
SET parameter_name = 'better_name';

SAVE GENERATED SKILL;

What's Next?¶

Improve Skills

Enhance existing skills with AI.

Improve Skills
Recommendations

Get AI-powered suggestions.

Recommendations