Building Workflows With Skills

Individual skills are useful. Chained together into workflows, they’re transformative. This lesson shows you how to design workflows that handle end-to-end processes — from raw input to finished output, with minimal supervision.

🔄 Quick Recall: From the OpenClaw lesson, you learned to vet community skills. For workflows, you’re combining skills you’ve built or vetted into larger processes. The same quality standards apply to every component.

The Workflow Architecture

A workflow is a sequence of skills that pass output forward:

Input → Skill A → Output A → Skill B → Output B → Skill C → Final Output

Each skill does one focused job. The workflow coordinates the handoffs.

Here’s a real example — a weekly client report workflow:

Raw Data (CSV) → /data-cleaner → Clean Data → /data-analyzer → Insights
→ /report-writer → Draft Report → /editor → Final Report

Four skills. One workflow. What used to take 90 minutes happens in 10.

Designing Your First Workflow

Step 1: Map the Manual Process

Before automating, write down exactly what you do. Here’s a content pipeline example:

Manual process:
1. Research a topic (30 min) — search, read articles, take notes
2. Create an outline (15 min) — structure the key points
3. Write the draft (45 min) — flesh out each section
4. Edit for clarity (20 min) — trim, sharpen, restructure
5. Format and publish (10 min) — add headers, links, images

Total: ~2 hours per piece

Step 2: Identify Skills for Each Step

Step 3: Define the Handoffs

Each skill’s output must be a clean input for the next skill. This is where most workflows break.

# Content Pipeline Workflow

## Phase 1: Research
Use /research-compiler
- Input: Topic keyword and 3-5 seed URLs
- Output: research-notes.md with summarized findings, key statistics, and source list
- Checkpoint: Review notes before proceeding

## Phase 2: Outline
Use /content-outliner
- Input: research-notes.md from Phase 1
- Output: content-outline.md with H2/H3 structure and key points per section
- Checkpoint: Approve structure before writing

## Phase 3: Draft
Use /content-drafter
- Input: content-outline.md + research-notes.md
- Output: draft-v1.md — full draft following the outline
- Checkpoint: Read draft, note major issues

## Phase 4: Edit
Use /content-editor
- Input: draft-v1.md + any revision notes from Phase 3 checkpoint
- Output: draft-final.md — polished version
- Checkpoint: Final review

## Phase 5: Format
Use /content-formatter
- Input: draft-final.md
- Output: published-content.md — ready to post

✅ Quick Check: Why are checkpoints between phases important in a workflow? (They let you catch problems early before errors cascade to later steps.)

The Master Prompt Pattern

You can orchestrate a full workflow with one prompt that references multiple skills:

I need to create a blog post about [topic].

Run this workflow:

1. First, use the research-compiler skill to gather information
   from these sources: [URLs]. Save findings to research-notes.md.
   STOP and show me the key findings before proceeding.

2. Then use the content-outliner skill to create an outline from
   the research notes. Save to content-outline.md.
   STOP and let me approve the structure.

3. Use the content-drafter skill to write the full post from the
   outline. Target 1,500 words. Save to draft.md.

4. Finally, use the content-editor skill to polish the draft.
   Apply our brand voice from brand-voice.md.
   Save final version to blog-post-final.md.

One prompt, four skills, checkpoints at critical transitions. Cowork runs each step sequentially and pauses where you told it to.

Scheduled Workflows

For workflows that run on a clock, Cowork’s scheduling feature removes the manual trigger:

Daily standup summary — Every morning at 8 AM:

1. Pull yesterday’s standup messages from Slack
2. Compile into a formatted summary
3. Save to /daily-summaries/YYYY-MM-DD.md

Weekly report — Every Friday at 4 PM:

1. Read all daily summaries from this week
2. Pull metrics from the project tracker
3. Generate the weekly status report
4. Draft an email to the team lead

Monthly newsletter — First Monday of each month:

1. Compile the month’s top content
2. Pull engagement metrics
3. Generate newsletter draft
4. Save to /newsletters/YYYY-MM.md

Set these up once. They run automatically. You review the output and send.

Error Handling in Workflows

Workflows break. The difference between amateur and production workflows is how they handle failure.

Add error handling instructions to your workflow skills:

## Error Handling

If a source file is missing:
- Log the missing file name
- Continue with available data
- Note in the output: "[Filename] was not available — this section may be incomplete"

If data quality issues are found:
- Flag specific issues in the output
- Don't guess at corrections — mark them for manual review
- Continue processing remaining data

If a connector is unavailable:
- Fall back to local files if available
- Note which external source was unreachable
- Suggest running the workflow again when the connector is restored

This turns “workflow failed” into “workflow completed with noted issues” — which is almost always more useful.

Testing Workflows End-to-End

Test with real data — not toy examples. Synthetic data hides real-world edge cases.

Run the workflow 5 times — consistency matters more than one successful run. If it works 3 out of 5 times, you have a reliability problem.

Time each phase — know where the bottlenecks are. If Phase 3 (drafting) takes 80% of the total time, that’s where optimization efforts should go.

Test the failure paths — deliberately break inputs. What happens when a file is missing? When data is malformed? When a connector is down?

✅ Quick Check: How many times should you test a workflow before considering it production-ready?

Workflow Documentation

Document your workflows so your team can use and maintain them:

# Content Pipeline Workflow

## Purpose
Produces blog posts from topic idea to published content.

## Skills Used
1. /research-compiler
2. /content-outliner
3. /content-drafter
4. /content-editor

## Required Inputs
- Topic keyword
- 3-5 source URLs (optional but recommended)
- brand-voice.md must be in the working folder

## Expected Output
- blog-post-final.md — ready to publish

## Estimated Time
- 15-20 minutes (Cowork time)
- 10 minutes (your review time at checkpoints)

## Known Issues
- Research phase occasionally includes outdated statistics — verify dates
- Editor skill sometimes over-trims technical content — review carefully for completeness

This documentation lives alongside your skills. When someone else on your team wants to use the workflow, they know exactly what to expect.

Key Takeaways

• Workflows chain skills: Input → Skill A → Skill B → Output
• Map your manual process first, then identify a skill for each step
• Clean handoffs between skills are critical — define output formats explicitly
• Use the master prompt pattern: one prompt, multiple skills, checkpoints
• Scheduled workflows automate recurring tasks hands-off
• Add error handling so workflows degrade gracefully instead of failing completely
• Test 5+ times with real data before considering a workflow production-ready

Up Next

Final lesson — the capstone. You’ll build a complete plugin with multiple skills, wire it together as a workflow, and prepare it for sharing. Everything you’ve learned in one package.