← ClaudeAtlas

brain-recalllisted

Search Symbiosis Brain for relevant context BEFORE acting. Trigger silently on these signals: (a) before first Edit/Write in an unfamiliar file or scope; (b) before grep across >3 files / a whole subtree; (c) when the user describes a bug or unexpected behavior — recall similar fix history; (d) after any subagent error or contradicted hypothesis — recall if this mistake is already documented; (e) before using an API / library that hasn't appeared in this session yet. Don't ask "should I check?" — just check. Report only when something relevant was found.
Krill113/symbiosis-brain · ★ 0 · AI & Automation · score 72
Install: claude install-skill Krill113/symbiosis-brain
# Brain Recall — Task Context Search ## Automatic action-trigger recall As of 2026-05-13, brain recall fires automatically before decision-class tool calls (Task, Edit, Write, MultiEdit, NotebookEdit, and whitelisted Bash commands like `git commit`, `pip install`, `docker push`). You will see a `[recall: N hits for "..."]` block in `<system-reminder>` injected by `hooks/brain-pre-action-trigger.sh`. This skill is still useful for: - Explicit recall when no tool call is imminent (pure exploration / discussion). - Searching by topic when the auto-hook query (derived from tool args) is too narrow. Runtime config: `~/.claude/symbiosis-brain-pre-action.json` (toggle matchers, edit Bash whitelist, etc.). Kill-switch: `SYMBIOSIS_BRAIN_PRE_ACTION_DISABLED=1`. Design: `decisions/2026-05-13-brain-recall-action-trigger-design.md` (in vault). ## When to use — concrete triggers Trigger silently on **any** of these signals (no permission needed, no narration if nothing relevant): 1. **Before first Edit/Write in a new file or scope.** Recall existing patterns + mistakes for that area before touching code. 2. **Before grep across >3 files or a whole subtree.** Memory may already have the answer — saves the read chain. 3. **When the user describes a bug, unexpected behavior, or "что-то не так".** Recall similar fix history before forming a hypothesis. 4. **After a subagent error or a contradicted hypothesis.** Recall whether this exact mistake is already documented as `mistakes/*`. 5.