Memory Recall

Automatically retrieve relevant memories from your openGauss-based memory store. This skill runs in a forked subagent context, keeping all intermediate search operations isolated from the main conversation.

When to Use

Claude will automatically invoke this skill when it determines that your question could benefit from historical context. You can also manually trigger it with /memory-recall <query>.

What It Does

This skill performs a three-layer progressive disclosure:

1. L1: Search - Searches for relevant memories using semantic vector search
2. L2: Expand - Retrieves full markdown sections for promising results
3. L3: Summarize - Curates and summarizes the findings

Only the final curated summary is returned to the main conversation, keeping your context window clean.

Progressive Disclosure

The skill autonomously evaluates search results and decides whether to drill deeper:

# L1: Search for relevant memories
$ openclaw ogmemory search "your query" --top-k 5

# L2: Expand promising results (full context)
$ openclaw ogmemory expand <chunk_id>

# L3: Return curated summary to main agent

Examples

Automatic invocation:

User: What did we decide about the caching strategy?
[Claude automatically invokes memory-recall skill]
→ Finds relevant discussions about caching
→ Returns summary of decisions and implementations

Manual invocation:

/memory-recall database configuration
→ Searches and retrieves all context about DB setup

Implementation Notes

• Runs in context: fork subagent
• Uses openGauss vector similarity search
• Filters by relevance score
• Expands top results for full context
• Returns concise summary to main conversation