AgentSkillsCN

recall

在代理的长期记忆中（Cortical Stack/Qdrant）搜索精炼的概念、过往的思考，以及经过整合的知识。当用户询问存储于向量数据库中的过往工作、决策，或特定领域的专业知识——而这些内容并未直接收录在Markdown文件中时，可使用此功能。

SKILL.md
--- frontmatter
name: recall
description: Search the agent's long-term memory (Cortical Stack/Qdrant) for refined concepts, past thoughts, and synthesized knowledge. Use this skill when the user asks about prior work, decisions, or specialized knowledge stored in the vector database that isn't immediately available in Markdown files.

Recall Skill (Cortical Stack)

Search the long-term memory stored in the local Qdrant instance. This skill uses the nomic-embed-text model via Ollama to perform semantic searches against the openclaw_memory collection.

Quick Start

Run the recall script with your search query:

bash
./skills/recall/scripts/recall.sh "your search query"

When to Use

  • •When asked "What do you remember about X?"
  • •To verify architectural decisions made in previous sessions.
  • •To retrieve snippets of articles or thoughts that were indexed by the Cortex.
  • •When memory_search (Markdown-only) doesn't return enough depth.

Implementation Details

The skill executes a two-step process:

  1. •Embedding: Generates a vector for the query using nomic-embed-text via the local Ollama API.
  2. •Vector Search: Queries the local Qdrant database for the top 5 most semantically similar points with a score threshold.

Troubleshooting

  • •Ollama Error: Ensure the Ollama service is running on port 11434 and the nomic-embed-text model is installed.
  • •Qdrant Error: Ensure the Qdrant container/service is running on port 6333.