← ClaudeAtlas

save-sessionlisted

Promote THIS CHAT's learnings into permanent memory. Scans ONLY the current conversation (plus this session's own staged blocks), categorizes, dedupes against the existing corpus by search, and writes memories/contexts/decisions/lessons/capabilities — each new reference/project memory gets recall cues so it is findable later. Surfaces only current-chat items; never pulls other chats' backlog into the chat (that belongs to a separate periodic maintenance pass). Default = review table + one approval pop-up. Pass `all` (or `auto` / `yolo` / `save all`) to skip the question and auto-write the provable-safe set, holding sensitive/secret/ambiguous items for a quick interactive pass. Distinct from a handoff brief (one-time, for the next chat) and from periodic memory maintenance (consolidation + backlog drain). User-invocable only.
donatassimkus/claude-ai-skills · ★ 1 · AI & Automation · score 77
Install: claude install-skill donatassimkus/claude-ai-skills
# /save-session User-invokable end-of-session graduation skill. Scans the current conversation, identifies promotable learnings, categorizes them, and writes the keepers into permanent state — the memory files, decisions log, lessons, contexts, and capability inventory that make up your persistent knowledge base. **Machine config.** Everything specific to ONE person's memory architecture belongs in a small companion config file, not in this method: the memory targets (which files to read and write), the scope-detection keyword table, the per-category output paths, and the sensitivity (surfacing-restricted content) handling. Write that config once for your own setup and keep it beside this skill; the steps below are the universal method that reads it: scan the current chat, categorize, scope-route, dedupe, write, index, summarize. **If no such config exists yet, the method still runs.** It promotes learnings generically to whatever local memory store the machine has, routes scope from the generic categories below, and exposes NO sensitivity-detection logic (it simply treats nothing as restricted). **Never invent a sensitivity rule the config did not declare** — a guessed rule either leaks what should have been held or holds everything and makes the skill useless. ## Invocation modes The argument decides the mode: | You type | Mode | Behavior | |---|---|---| | no argument | **Interactive** (default) | Scan → present a table → one `AskUserQuestion` approval → write approve