From Harness Base
Store and retrieve durable facts from the Mem0 vector-memory MCP. Use at session start to recall task-relevant memories, and whenever you learn a durable user- or domain-fact worth keeping.
How this skill is triggered — by the user, by Claude, or both
Slash command
/harness-base:memory-vector-storeThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
A Mem0 MCP server provides semantic memory for knowledge too large to hold in
A Mem0 MCP server provides semantic memory for knowledge too large to hold in
context. This skill covers when and how to use it. It does not replace
file-memory — CLAUDE.md and .claude/memory/*.md still hold code-shaped
knowledge you grep and review.
Query the memory MCP with a natural-language description of the current task before acting. The server returns the most semantically similar memories.
Add a memory via the MCP when you learn a fact a future session would want:
When storing:
.claude/memory/*.md, secrets go nowhere.CLAUDE.md..claude/memory/*.md.When in doubt, prefer file-memory: it is auditable and git-diffable. Reach for the vector store only for the large, fuzzy corpus it is built for.
This skill needs the Mem0 MCP server. It is not auto-started by the plugin —
add it to your project's .mcp.json only if you want vector memory, then set
MEM0_API_KEY in your environment:
{
"mcpServers": {
"mem0": {
"command": "npx",
"args": ["-y", "@mem0/mcp-server"],
"env": { "MEM0_API_KEY": "${MEM0_API_KEY}" }
}
}
}
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub camilool8/harness-engineering-templates --plugin harness-base