Agent Skills: Managing Temporary Scripts

Managing Temporary Scripts

I help you create and execute temporary scripts during workflow execution. Perfect for API integrations, data processing with specialized libraries, and creating temporary tools that execute and return results.

When I Activate

I automatically activate when you:

• Need to interact with external APIs (Reddit, Twitter, GitHub, etc.)
• Want to use specific libraries not available in Claude Code
• Need to process data with custom code
• Ask "how do I call an API in a workflow?"
• Mention "temporary script", "execute code", "API client"
• Need credentials/API keys for external services

Key Concept

Temporary scripts are code files that:

1. Are created during workflow execution
2. Execute via Bash tool
3. Return results to workflow
4. Are automatically cleaned up after workflow completion

Supported Languages:

• Python (with pip packages)
• Node.js/JavaScript (with npm packages)
• Shell/Bash scripts
• Ruby, Go, or any executable language

Quick Example

# 1. Ask for API credentials
AskUserQuestion:"Reddit API key needed":api_key ->

# 2. Create Python script with embedded credentials
general-purpose:"Create Python script: reddit_client.py with {api_key}":script_path ->

# 3. Execute script and capture output
Bash:"python3 {script_path}":reddit_data ->

# 4. Process results in workflow
general-purpose:"Analyze {reddit_data} and create summary":analysis ->

# 5. Cleanup happens automatically

Script Lifecycle

See script-lifecycle.md for complete details.

Overview:

1. Creation
   ↓
   Write script to /tmp/workflow-scripts/

2. Preparation
   ↓
   Set permissions (chmod +x)
   Install dependencies if needed

3. Execution
   ↓
   Run via Bash tool
   Capture stdout/stderr

4. Data Return
   ↓
   Parse output (JSON, CSV, text)
   Pass to next workflow step

5. Cleanup
   ↓
   Remove script files
   Clean temp directories

Common Use Cases

1. API Integration

Reddit API Client:

# /tmp/workflow-scripts/reddit_client.py
import requests
import json
import sys

api_key = sys.argv[1]
subreddit = sys.argv[2]

headers = {'Authorization': f'Bearer {api_key}'}
response = requests.get(
    f'https://oauth.reddit.com/r/{subreddit}/hot.json',
    headers=headers
)

print(json.dumps(response.json(), indent=2))

In workflow:

$script-creator:"Create reddit_client.py":script ->
Bash:"python3 {script} {api_key} programming":posts ->
general-purpose:"Parse {posts} and extract top 10 titles"

2. Data Processing

CSV Analysis:

# /tmp/workflow-scripts/analyze_data.py
import pandas as pd
import sys

df = pd.read_csv(sys.argv[1])
summary = df.describe().to_json()
print(summary)

In workflow:

general-purpose:"Create analyze_data.py script":script ->
Bash:"pip install pandas && python3 {script} data.csv":analysis ->
general-purpose:"Interpret {analysis} and create report"

3. Web Scraping

Article Scraper:

// /tmp/workflow-scripts/scraper.js
const axios = require('axios');
const cheerio = require('cheerio');

async function scrapeArticles(url) {
  const {data} = await axios.get(url);
  const $ = cheerio.load(data);

  const articles = [];
  $('.article').each((i, el) => {
    articles.push({
      title: $(el).find('.title').text(),
      url: $(el).find('a').attr('href')
    });
  });

  console.log(JSON.stringify(articles));
}

scrapeArticles(process.argv[2]);

In workflow:

general-purpose:"Create scraper.js":script ->
Bash:"npm install axios cheerio && node {script} https://news.site":articles ->
general-purpose:"Process {articles}"

Script Templates

See script-templates.md for complete library.

Quick templates:

• API Client (REST, GraphQL)
• Data Processing (CSV, JSON, XML)
• Web Scraping (HTML parsing)
• File Processing (PDF, images, documents)
• Database Access (PostgreSQL, MySQL, MongoDB)
• Message Queues (RabbitMQ, Kafka)
• Cloud Services (AWS S3, GCS, Azure)

Security Best Practices

See security.md for comprehensive guide.

Quick checklist:

✅ Credentials Management:

• Pass via command-line arguments (not hardcoded)
• Use environment variables for sensitive data
• Clean up after execution

✅ File Permissions:

chmod 700 /tmp/workflow-scripts/script.py  # Owner only

✅ Output Sanitization:

• Validate script output before using
• Escape special characters
• Limit output size

✅ Dependency Management:

• Use virtual environments for Python
• Specify exact package versions
• Avoid running arbitrary code

Integration with Workflows

See integration-patterns.md for detailed patterns.

Pattern 1: Simple Script Execution

general-purpose:"Create script.py that fetches data":script ->
Bash:"python3 {script}":data ->
general-purpose:"Process {data}"

Pattern 2: Script with User Input

AskUserQuestion:"API credentials needed":creds ->
general-purpose:"Create api_client.py with {creds}":script ->
Bash:"python3 {script}":results ->
general-purpose:"Analyze {results}"

Pattern 3: Parallel Script Execution

general-purpose:"Create multiple API clients":scripts ->
[
  Bash:"python3 {scripts.reddit}":reddit_data ||
  Bash:"python3 {scripts.twitter}":twitter_data ||
  Bash:"python3 {scripts.github}":github_data
] ->
general-purpose:"Merge all data sources"

Pattern 4: Iterative Processing

@process_batch ->
general-purpose:"Create batch_processor.py":script ->
Bash:"python3 {script} batch_{n}":results ->
(if results.has_more)~> @process_batch ~>
(if results.complete)~> general-purpose:"Finalize"

Script Directory Structure

/tmp/workflow-scripts/
├── {workflow-id}/              # Unique per workflow
│   ├── reddit_client.py
│   ├── data_processor.py
│   ├── requirements.txt        # Python dependencies
│   ├── package.json            # Node.js dependencies
│   └── .env                    # Environment variables

Automatic cleanup after workflow:

rm -rf /tmp/workflow-scripts/{workflow-id}

Creating Scripts in Workflows

Method 1: Inline Script Creation

general-purpose:"Create Python script:
```python
import requests
import sys

api_key = sys.argv[1]
response = requests.get(
    'https://api.example.com/data',
    headers={'Authorization': f'Bearer {api_key}'}
)
print(response.text)

Save to /tmp/workflow-scripts/api_client.py":script_path ->

Bash:"python3 {script_path} {api_key}":data

### Method 2: Template-Based Creation

```flow
general-purpose:"Use template: api-rest-client
- Language: Python
- API: Reddit
- Auth: Bearer token
- Output: JSON
Create script in /tmp/workflow-scripts/":script ->

Bash:"python3 {script}":data

Method 3: Multi-File Scripts

general-purpose:"Create script package:
- main.py (entry point)
- utils.py (helper functions)
- requirements.txt (dependencies)
Save to /tmp/workflow-scripts/package/":package_path ->

Bash:"cd {package_path} && pip install -r requirements.txt && python3 main.py":data

Dependency Management

Python (pip)

general-purpose:"Create requirements.txt:
requests==2.31.0
pandas==2.0.0
Save to /tmp/workflow-scripts/":deps ->

general-purpose:"Create script.py":script ->

Bash:"pip install -r {deps} && python3 {script}":data

Node.js (npm)

general-purpose:"Create package.json with dependencies":package ->

general-purpose:"Create script.js":script ->

Bash:"cd /tmp/workflow-scripts && npm install && node {script}":data

Virtual Environments

Bash:"python3 -m venv /tmp/workflow-scripts/venv":venv ->

general-purpose:"Create script in venv":script ->

Bash:"source /tmp/workflow-scripts/venv/bin/activate && python3 {script}":data

Error Handling

Capturing Errors

Bash:"python3 {script} 2>&1":output ->

(if output.contains('Error'))~>
  general-purpose:"Parse error: {output}":error ->
  @review-error:"Script failed: {error}" ~>

(if output.success)~>
  general-purpose:"Process {output}"

Retry Logic

@retry ->
Bash:"python3 {script}":result ->

(if result.failed)~>
  general-purpose:"Wait 5 seconds" ->
  @retry ~>

(if result.success)~>
  general-purpose:"Process {result}"

Output Formats

Scripts can return data in various formats:

JSON (Recommended)

import json
result = {"data": [...], "status": "success"}
print(json.dumps(result))

CSV

import csv
import sys
writer = csv.writer(sys.stdout)
writer.writerows(data)

Plain Text

for item in results:
    print(f"{item['title']}: {item['url']}")

Cleanup

Automatic Cleanup

Cleanup happens automatically after workflow completion:

# At end of workflow execution:
general-purpose:"Remove all scripts in /tmp/workflow-scripts/{workflow-id}"

Manual Cleanup

For long-running workflows:

general-purpose:"Create and execute script":result ->
general-purpose:"Process {result}":output ->
Bash:"rm -rf /tmp/workflow-scripts/{script-dir}":cleanup

Best Practices

DO:

✅ Use unique workflow IDs for script directories ✅ Pass credentials as arguments, not hardcoded ✅ Validate and sanitize script output ✅ Use virtual environments for Python ✅ Specify exact dependency versions ✅ Return structured data (JSON preferred) ✅ Clean up after workflow completion ✅ Set restrictive file permissions ✅ Use timeouts for script execution ✅ Log script output for debugging

DON'T:

❌ Hardcode API keys in scripts ❌ Execute untrusted code ❌ Store sensitive data in script files ❌ Leave scripts after workflow ❌ Use global Python/Node packages ❌ Ignore script errors ❌ Return massive outputs (>1MB) ❌ Use system-wide directories

Tips for Effective Script Management

1. Use Descriptive Names: reddit_api_client.py not script.py

2. Return Structured Data: Always use JSON when possible

3. Error Messages: Include detailed error messages in output

4. Logging: Add logging for debugging:

import logging
logging.basicConfig(level=logging.INFO)
logging.info(f"Fetching data from {url}")

5. Validation: Validate inputs before execution

6. Timeouts: Set execution timeouts:

Bash:"timeout 30 python3 {script}":data

Related Skills

• creating-workflows: Create workflows that use temp scripts
• executing-workflows: Execute workflows with script steps
• managing-agents: Temp agents vs temp scripts
• debugging-workflows: Debug script execution issues

Advanced Topics

See detail files for:

• script-lifecycle.md - Complete lifecycle management
• script-templates.md - 20+ ready-to-use templates
• security.md - Security best practices
• integration-patterns.md - Workflow integration patterns

Ready to use temporary scripts? Just describe what API or processing you need!