hindsight-memorylisted

Interact with and integrate Hindsight long-term AI memory in Claude Code via the `hindsight` CLI. Use for recalling past context, reflecting over a memory bank, retaining facts/decisions, managing banks/entities/mental-models, and understanding the auto-recall/auto-retain hooks this plugin installs. Trigger phrases: "what do I remember about", "recall from my memory", "reflect on my past work", "retain this", "save to my memory bank", "hindsight memory", "check my long-term memory", "is memory working". NOT for short-term session context (Claude already has the transcript), NOT for code/structural search (use CodeGraph/Semble/grep), NOT for building a memory framework from scratch (use the memory-systems skill).
viktorbezdek/skillstack · ★ 9 · AI & Automation · score 77

Install: claude install-skill viktorbezdek/skillstack

# Hindsight Memory Hindsight is a semantic long-term memory system. This plugin wires Claude Code to a Hindsight server (external API) so that **relevant memories are recalled before every prompt** and **the conversation is retained after every turn** — and gives you a vocabulary for driving that memory by hand when you want to. The transport is the installed `hindsight` CLI (verify with `hindsight version`). It already knows the API endpoint from `HINDSIGHT_API_URL` / `HINDSIGHT_API_KEY` or a named profile, so every command below "just works" against your server. ## How the automatic memory works Three hooks (in `hooks/hooks.json`) run with no action from you: | Hook | Event | Script | What it does | |------|-------|--------|--------------| | Recall | `UserPromptSubmit` | `scripts/recall.py` | Queries the bank for memories relevant to your prompt and injects them as an invisible `<hindsight_memories>` block — Claude sees them, the chat does not. | | Retain | `Stop` (async) | `scripts/retain.py` | Reads the transcript, **strips any injected `<hindsight_memories>` block**, and stores it under a deterministic per-session document id. | | Health | `SessionStart` | `scripts/session_start.py` | Silent reachability check; logs only with debug on. | Every hook **fails open**: if the server is slow or unreachable, the turn is never blocked and no error reaches you. The recall budget defaults to `mid` and the query is capped, so latency stays bounded. The single most important