What problem does it solve?
This Skill automates the tedious, manual process of reading Claude Code session logs to discover, abstract, and prioritize new skill ideas so maintainers can continuously grow a backlog of implementable trading skills without sifting through sessions by hand.
Core Features & Use Cases
- Session log mining: Enumerates and scans recent Claude Code session files to extract user messages, tool usage, and timing signals.
- Deterministic signal detection & deduplication: Detects skill-usage patterns, errors, repetitive workflows, automation requests, and unresolved messages, then deduplicates candidates against existing skills using Jaccard similarity.
- LLM abstraction and scoring: Invokes Claude CLI to abstract user requests into candidate ideas, scores novelty/feasibility/trading value, and merges results into a prioritized backlog for downstream skill generation.
- Use case: Run as a weekly pipeline to automatically convert a week's worth of developer sessions into a ranked list of implementable skill ideas for the trading skills repository.
Quick Start
Use the skill-idea-miner to scan the last seven days of Claude Code session logs, extract candidate ideas, score and deduplicate them, and append the prioritized entries to the skill generation backlog.