Cognitive Process Automation
skill-creator "I need to validate startup ideas by researching market size, analyzing competitors, and evaluating founder-market fit"
This creates a structured skill that runs your exact validation methodology every time, maintaining quality and consistency without manual effort.
Phase 1: Process Identification
- Identify a cognitive process you've done 10+ times manually
- Document the current steps (even if informal)
- Define what "good output" looks like for this process
Phase 2: Skill Creation
- Use skill-creator to generate initial structure
- Specify exact requirements and quality standards
- Include negative instructions (what NOT to do)
- Break complex processes into sequential phases
Phase 3: Iteration
- Run the skill on a real case
- Document gaps and improvements needed
- Update instructions based on output quality
- Repeat 3-4 times until output matches manual quality
Phase 4: Optimization
- Add constraints and edge case handling
- Refine phase transitions and data flow
- Document the skill for team sharing
- Publish if broadly applicable
Example 1: Code Review Skill Input: "Create a skill for conducting security-focused code reviews for our Python APIs" Output: Skill with phases for dependency analysis, authentication checks, input validation review, and vulnerability assessment - runs same standards every time
Example 2: Customer Research Skill Input: "Automate our pre-feature customer research process" Output: Skill that interviews users, analyzes usage data, checks competitor features, and produces structured recommendations before any new development
Example 3: Technical Writing Skill Input: "Standardize our API documentation process" Output: Skill that generates consistent docs with examples, error handling, authentication details, and testing instructions for every endpoint
Start with proven processes: Only automate workflows you've mastered manually. The skill encodes YOUR judgment.
Be hyper-specific: Instead of "research competitors," specify "find 5-8 direct competitors, extract pricing tiers, analyze G2 reviews for complaints, flag recent funding rounds."
Use negative constraints: "Do not sugarcoat results," "Do not skip financial analysis," "Do not present estimates as facts."
Design sequential phases: Break complex processes into steps where each phase produces inputs for the next. Better depth than trying to do everything at once.
Plan for evolution: Skills improve through use. Expect 3-4 iterations before solid performance, 10+ iterations before exceeding manual quality.
Automating unfamiliar processes: Don't create skills for workflows you haven't done successfully multiple times manually.
Vague quality standards: "Do good research" produces mediocre output. Specify exactly what thorough looks like.
Monolithic design: Single-phase skills produce shallow analysis. Break into logical sequential steps.
Set-and-forget mentality: Skills need iteration. Plan to improve based on real usage, not perfect first versions.
Hoarding useful skills: If your process solves common problems, publish it. Team skills multiply organizational capability.
Skipping documentation: Undocumented skills become unmaintainable. Include context, constraints, and evolution notes.