Find Hidden Workflows — AI Skill Discovery Engine

Turn your messy chat history into a library of automated agents. Most developers have hundreds of interactions hidden in their LLM logs that represent repeatable processes, but they lack the time to extract and formalize them. This skill scans your conversation exports or current context to identify recurring patterns, score them for automation suitability, and generate production-ready AI skill specifications.

It moves beyond simple keyword matching to look for task episodes. By analyzing the user goal, inputs, constraints, and outcomes across multiple interactions, it identifies which workflows are stable enough to be offloaded to a dedicated agent. Every candidate is evaluated against a weighted scoring model that considers recurrence evidence, process stability, and reusable value.

What it does

• Extract task episodes by recording goals, inputs, steps, and outcomes from raw chat data.
• Cluster patterns to group similar interactions into logical workflow candidates.
• Score candidates using a weighted 100-point scale across six dimensions like input clarity and automation suitability.
• Formalize specifications into a standard template including operating procedures, decision rules, and output schemas.
• Validate evidence by identifying missing data points and asking for user confirmation before assuming a pattern is established.

How it works

1. Define scope by identifying the supplied conversation sources and any privacy exclusions.
2. Analyze history to extract and cluster recurring task patterns and reusable process steps.
3. Review scoring to see which workflows have the highest recurrence and stability for automation.
4. Select candidates to convert into formal AI skill specifications.

Frameworks & tools

Works with standard conversation exports (JSON/Markdown), Claude interaction logs, and text-based chat history.

Why this beats prompting it yourself

Manual extraction is tedious and prone to recency bias. This skill uses a weighted scoring engine to objectively prove a workflow's value based on historical evidence rather than gut feeling.

Use cases

• Convert recurring PR review styles into a standardized code-review skill.
• Turn repeated documentation requests into a technical-writing agent.
• Extract complex data extraction patterns into a reusable parsing skill.
• Standardize ad-hoc troubleshooting steps into a diagnostic agent.

Known limitations

Cannot access external accounts or history not provided in the context. Scoring is provisional if based on limited evidence. Requires user validation for assumptions.