patrick-motard/extract-workflow

Extract Workflow

Analyze recent conversation to identify if a workflow should be automated as a skill, command, agent, or documentation.

Steps

1. Summarize the recent work

• Review the last 10-20 messages in the conversation
• Identify the complex task or workflow that was performed
• Extract the key steps, tools used, and decisions made
• Note any repeated patterns or manual processes

2. Analyze automation potential

• Is this a repeatable pattern?
• How frequently would this be used?
• What's the complexity level?
• Does it require multi-step coordination?
• Would automation save significant time/tokens?

3. Present options to user

Ask the user to select the appropriate automation type:

| Option | Description | |--------|-------------| | Skill | Model-invoked capability, agent decides when to use it, focused expertise | | Slash Command | Explicit prompt expansion, user-triggered, simple-to-moderate complexity | | Agent | Complex task needing dedicated context window and specialized tools | | AGENTS.md | Documentation/guidelines that should be in project or global AGENTS.md | | None | Not worth automating, one-off task |

4. For each option, explain

• Why this type fits (or doesn't fit)
• Trade-offs (e.g., slash commands require manual trigger, skills are auto-invoked)
• Suggested name and structure
• Key components that would be included

5. After user selects type, ask scope

• Current project only (.pi/skills/, .pi/prompts/, .pi/extensions/ in current repo)
• Global (~/.pi/agent/skills/, ~/.pi/agent/prompts/, ~/.pi/agent/extensions/)
• Both (create globally and optionally override in project)

6. If AGENTS.md selected, ask

• Project AGENTS.md (current repo only)
• Global AGENTS.md (~/.pi/agent/AGENTS.md - applies to all projects)
• Both (global guidelines with project-specific overrides)

7. Create the automation

• Generate the appropriate file structure
• Include clear documentation
• Add examples where relevant
• Follow best practices for the chosen type
• For skills: include proper frontmatter with name and description
• For prompts: use Markdown with {{variables}} for parameters
• For AGENTS.md: structure as clear guidelines with context

8. Confirm and show

• Display the file path created
• Show the content
• Explain how to use/invoke it
• Mention any next steps (e.g., testing, iteration)

Important considerations

• Skills are great for model-invoked capabilities with focused expertise
• Prompt templates are great for explicit, frequently-used prompts
• Extensions add tools, commands, UI — use for anything programmatic
• AGENTS.md is for persistent context/guidelines that should always be loaded
• Consider token costs: AGENTS.md loads every conversation, skills/prompts load on-demand