Building Custom Workflows

Learn how to create custom workflows that match your specific content processing needs.

Two Ways to Create Workflows

AI-Assisted (Recommended)
Manual

Use the build command to generate workflows from natural language:

looplia build "analyze podcast transcripts and create newsletter content"

The AI will:

Discover available skills
Match your requirements
Generate a valid workflow

See build command for details.

Create a workflow file directly in ~/.looplia/workflows/:

touch ~/.looplia/workflows/my-workflow.md

Then edit with your preferred editor.

Manual Workflow Creation

Create the workflow file

# Create a new workflow
cat > ~/.looplia/workflows/podcast-summary.md << 'EOF'
---
name: podcast-summary
version: 1.0.0
description: Summarize podcast transcripts into shareable content

steps:
  - id: analyze
    skill: media-reviewer
    mission: |
      Analyze the podcast transcript to extract:
      - Main topics discussed
      - Key quotes from speakers (with attribution)
      - Actionable takeaways for listeners
    input: ${{ sandbox }}/inputs/content.md
    output: ${{ sandbox }}/outputs/analysis.json

  - id: shareable
    skill: writing-kit-assembler
    mission: |
      Create shareable content from the analysis:
      - 3 tweet-sized quotes (under 280 chars)
      - LinkedIn post summary (100-200 words)
      - Newsletter blurb (50-100 words)
    needs: [analyze]
    input: ${{ steps.analyze.output }}
    output: ${{ sandbox }}/outputs/shareable.json
    final: true
---

# Podcast Summary Workflow

Transforms podcast transcripts into shareable social content.

## Usage

\`\`\`bash
looplia run podcast-summary --file transcript.md
\`\`\`
EOF

Test the workflow

# Create a test transcript
echo "Host: Welcome to the show. Guest: Thanks for having me..." > test-transcript.md

# Run your new workflow
looplia run podcast-summary --file test-transcript.md

Iterate and refine
- Check outputs in ~/.looplia/sandbox/*/outputs/
- Adjust missions for better results
- Add validation criteria

Workflow Template

Here’s a template for common patterns:

---
name: my-workflow
version: 1.0.0
description: Short description of what this workflow does

steps:
  # Step 1: Analyze input content
  - id: analyze
    skill: media-reviewer
    mission: |
      Analyze the content to extract:
      - [List specific requirements]
      - [Include quantities where applicable]
    input: ${{ sandbox }}/inputs/content.md
    output: ${{ sandbox }}/outputs/analysis.json
    validate:
      required_fields: [field1, field2]

  # Step 2: Transform/generate
  - id: transform
    skill: idea-synthesis    # or another skill
    mission: |
      Based on the analysis:
      - [Describe transformation]
      - [Specify output format]
    needs: [analyze]
    input: ${{ steps.analyze.output }}
    output: ${{ sandbox }}/outputs/transformed.json

  # Step 3: Assemble final output
  - id: assemble
    skill: writing-kit-assembler
    mission: |
      Combine inputs into final deliverable:
      - [Describe final format]
    needs: [analyze, transform]
    input:
      - ${{ steps.analyze.output }}
      - ${{ steps.transform.output }}
    output: ${{ sandbox }}/outputs/final.json
    final: true
---

# My Workflow

Description and usage instructions...

Available Skills

These skills are available in the default installation:

Content Analysis

Skill	Description
`media-reviewer`	Deep content analysis, theme extraction, quote identification

Content Generation

Skill	Description
`idea-synthesis`	Generate hooks, angles, questions, prompts
`writing-kit-assembler`	Combine inputs into structured output

Design Patterns

Sequential Processing

Steps run one after another:

steps:
  - id: step1
    skill: media-reviewer
    # ...

  - id: step2
    skill: idea-synthesis
    needs: [step1]
    input: ${{ steps.step1.output }}
    # ...

Fan-Out Pattern

One input feeds multiple parallel analyses:

steps:
  - id: analyze
    skill: media-reviewer
    input: ${{ sandbox }}/inputs/content.md
    output: ${{ sandbox }}/outputs/analysis.json

  - id: social-content
    skill: writing-kit-assembler
    needs: [analyze]
    mission: Create social media posts
    input: ${{ steps.analyze.output }}
    output: ${{ sandbox }}/outputs/social.json

  - id: newsletter
    skill: writing-kit-assembler
    needs: [analyze]
    mission: Create newsletter content
    input: ${{ steps.analyze.output }}
    output: ${{ sandbox }}/outputs/newsletter.json

Fan-In Pattern

Multiple analyses combine into one output:

steps:
  - id: themes
    skill: media-reviewer
    mission: Extract themes
    # ...

  - id: quotes
    skill: media-reviewer
    mission: Extract quotes
    # ...

  - id: combined
    skill: writing-kit-assembler
    needs: [themes, quotes]
    input:
      - ${{ steps.themes.output }}
      - ${{ steps.quotes.output }}
    # ...

Mission Writing Tips

Be Specific

# Vague (avoid)
mission: Analyze the content

# Specific (better)
mission: |
  Analyze the content to extract:
  - 3-5 main themes with supporting evidence
  - At least 5 quotable statements (under 280 characters each)
  - Key statistics or data points
  - Inferred target audience

Include Format Requirements

mission: |
  Generate social media content:
  - 5 tweets (each under 280 characters, with hashtags)
  - 1 LinkedIn post (150-200 words, professional tone)
  - 1 Instagram caption (casual tone, with emoji suggestions)

Provide Context

mission: |
  Analyze this podcast transcript for a technology newsletter audience.
  Focus on:
  - Practical implications for developers
  - Controversial or surprising claims
  - Quotable insights from the guest

Validation

Add validation to ensure quality:

- id: analyze
  skill: media-reviewer
  validate:
    required_fields: [themes, quotes, keyPoints]
    min_quotes: 3
    min_key_points: 5

If validation fails, the step is marked incomplete and you can adjust and retry.

Debugging

Check Outputs

# View step outputs
cat ~/.looplia/sandbox/*/outputs/analysis.json | jq

# Check validation state
cat ~/.looplia/sandbox/*/validation.json

View Logs

# Session logs
cat ~/.looplia/sandbox/*/logs/session.log

Mock Mode

Test workflow structure without API calls:

looplia run my-workflow --file test.md --mock