Stats
Actions
Tags
Help us improve
Share bugs, ideas, or general feedback.
From ai-memory
Deep recovery from a previous session — read session content and resume work.
npx claudepluginhub dankinsoid/ai-memory --plugin ai-memoryHow this skill is triggered — by the user, by Claude, or both
Slash command
/ai-memory:loadThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Parse ARGUMENTS to determine which session to load:
Restores previous Claude Code session context by reading transcripts directly, avoiding LLM calls and token costs.
Presents menu-driven interface to scan, select, and resume saved sessions from CONTEXT files. Extracts headers, summaries next tasks, hot files; supports active/archived.
Revives previous sessions from YAML squirrel logs and transcripts to reconstruct context. Browses recent activity across walnuts, filters by topic/person/timeframe, flags unsaved sessions.
Share bugs, ideas, or general feedback.
Parse ARGUMENTS to determine which session to load:
No args → present the recent sessions from SessionStart context for the user to choose.
Show 3-4 sessions using AskUserQuestion with options (label = title, description = summary + date).
Include an "Other / search" option so the user can type a query if none match.
If "Other" is picked, ask for a search query and proceed to step 3.
"last" / "latest" → load the most recent session directly.
Call memory_search with tags: ["session", "project/<project>"], limit: 1,
exclude_session_id: <current session_id>. Load the result directly.
Free text → first scan the sessions already in SessionStart context.
If a session's title or summary clearly matches the user's request, use its [[ref]] directly —
no need to call memory_search.
If no context match, fall back to memory_search:
tags: ["session", "project/<project>"], exclude_session_id: <current session_id>since/until (resolve relative to today)query (translate to English)If memory_search was used with query or topic-based tags (fuzzy/ambiguous),
always let the user pick using AskUserQuestion with up to 4 candidates as options
(label = title, description = summary). The user can pick "Other" to refine.
If the match was exact (from context or limit: 1 with no query), load directly.
Once a session is identified, call memory_load_session with its ref (the [[ref]] wikilink stem).
This returns compact notes + transcript tail — optimized for recovery.
Do NOT announce what you loaded. Do not say "here's what I recovered" or "what should we do next?".
Instead: