一键导入
recall
Load saved knowledge from ~/.claude/learnings/ into the current session context.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Load saved knowledge from ~/.claude/learnings/ into the current session context.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Deep research a topic using parallel agents, then synthesize and save comprehensive knowledge to ~/.claude/learnings/.
Delete saved knowledge from ~/.claude/learnings/ and optionally clean up knowledge graph.
Research a topic and save structured knowledge to ~/.claude/learnings/ for use in future sessions.
| name | recall |
| description | Load saved knowledge from ~/.claude/learnings/ into the current session context. |
| allowed-tools | ["Bash","Read","Glob","Grep","mcp__memory__search_nodes"] |
| argument-hint | <topic> - full name, slug, or partial match. Omit to list all. |
Load saved knowledge from ~/.claude/learnings/ into the current session context.
Argument: $ARGUMENTS
If empty, list all available knowledge files:
ls ~/.claude/learnings/*.md 2>/dev/null
Display as a list with topic names extracted from frontmatter. Stop here.
Search ~/.claude/learnings/ in this order:
~/.claude/learnings/<slug>.md~/.claude/learnings/*<query>*.mdmcp__memory__search_nodes available): search for entities of type "learning"If multiple matches, list them and ask the user to pick one.
If no matches, tell the user: no knowledge found, suggest /learn <topic>.
Read the matching file. Absorb the content into your working context. Use it to inform your responses for the rest of the session.
Do NOT recite the entire file back to the user. Instead, confirm briefly:
Loaded knowledge on "" (verified , confidence: ).
If the user asks for details, then share relevant sections.
If last_verified is >90 days old, note it once:
Note: Last verified . Will cross-check key details as needed.
If you discover outdated content during the session, silently update the file (fix content, bump last_verified, adjust confidence). No user confirmation needed for routine maintenance.