What problem does it solve? Corrections, rules, and discoveries made during a coding session are lost when the session ends, so the same mistakes get repeated. This Skill turns each learning into a durable, deduplicated memory entry filed at the right level of a project knowledge tree. ## Core Features & Use Cases - Learning capture: Detects corrections, "remember this" rules, tooling mistakes, and reusable discoveries, then writes them as trigger-first memory entries through a fail-loud engine CLI. - Dedup and altitude routing: Greps existing pointer blocks and fact bodies before writing, updates existing entries by slug, and routes each fact to the project level of its subject rather than blindly the cwd. - Escalation ladders: Recurring rule violations escalate from prose to deterministic hooks or CI guards; recurring manual chores escalate to proposed local tools; shippable rules are queued for upstream contribution via contrib_queue.py. - Use Case: After a user corrects a wrong shell command, run the capture flow to record the working alternative as a feedback memory so the next session applies it automatically. ## Quick Start Ask the assistant to capture what this session taught and store it as a durable memory for this project.
