IMA Knowledge Distiller

Problem

When notes, documents, web pages, and knowledge bases grow over time, reusable know-how often stays fragmented. Asking a model to summarize it can produce generic bullets, while hand-writing a Skill requires repeated work on scope, structure, references, and validation. IMA Knowledge Distiller addresses this by converting IMA knowledge base content into structured, actionable skill instructions.

Core Workflow

The Skill runs in five stages:
- Discovery: lists candidates with search_knowledge_base, get_knowledge_list, or search_knowledge, then lets the user choose what to distill.
- Acquisition: downloads files, reads note bodies, and fetches web pages into a temporary workspace.
- Distillation: extracts themes, methodology, decision logic, pitfalls, and metrics, then recommends a method, mindset, mixed, data-driven, or tool-oriented structure.
- Generation: creates the Skill directory, writes SKILL.md and references/, and keeps the main file focused.
- Validation: checks trigger conditions, executability, length, sensitive data, Gotchas, reference integrity, and naming rules.

Boundaries

It depends on the ima Skill for knowledge base access, and may need PDF, Word, Excel, or web-fetch support depending on source types. The output is intended as structured knowledge for agents or engineers, not marketing copy. Before running it, confirm the source scope and avoid distilling internal IDs, credentials, or customer data into the final Skill.