neuroxlisted

# Neurox Memory Skill Neurox is not a note-taking tool. It is the agent's persistent memory. > Remember first. Investigate second. Act third. Learn before closing. --- ## 1. Session Protocol Every work session follows this loop: ``` START ─→ RECALL ─→ INVESTIGATE ─→ WORK ─→ LEARN ─→ CLOSE ``` ### Start 1. `session_start(title, directory, branch, namespace)` 2. `context(namespace)` — load project memory 3. `context(namespace, files=...)` — load file-linked memory if you know the target files 4. Read returned context before proceeding ### Recall Before answering, proposing, or editing: - `recall(query, namespace)` with short keyword queries - use filters: `observation_type`, `kind`, `files`, `include_stale` - prefer multiple targeted passes over one vague query ### Investigate Memory gives priors. Files give truth. Before making claims about code: - open the target files - read nearby tests - check schema/config when relevant - search for similar patterns in the codebase - compare with prior decisions in Neurox ### Work Act with the full context of memory + investigation. ### Learn After meaningful work, save durable learnings: - `save` new knowledge - `update` refined knowledge - `invalidate` wrong knowledge ### Close `session_end` with structured summary: - `Goal:` what you set out to do - `Discoveries:` what you learned - `Accomplished:` what got done - `Next:` what remains --- ## 2. Three Reflexes ### Reflex A: Remember the user Before coding, recove