Agent Skills open standard
Engra ships a cross-agent SKILL.md verified on Cursor, Claude Code, OpenClaw, VS Code Copilot, Windsurf, Cline, and OpenCode. Pair it with Memory MCP for a durable memory workflow.
Integration versions v1.9.0
- REST API
- v1.6.0
- MCP
- v1.9.0
- Agent Skill
- v1.9.0
- Memory engine
- v0.5.0
Declared in skills/engra-memory/SKILL.md frontmatter (version field).
What is an Agent Skill?
Agent Skills follow the open Agent Skills standard: YAML frontmatter (name, description) plus markdown instructions. Hosts match skills by description and load them on demand. The same SKILL.md works across many coding agents; Engra Memory also requires Memory MCP in your client to call tools like memory_search.
Recommended (Skill + MCP)
| Agent | Personal | Project |
|---|---|---|
| Cursor | ~/.cursor/skills/ | .cursor/skills/ |
| Claude Code | ~/.claude/skills/ | .claude/skills/ |
| OpenClaw | ~/.openclaw/skills/ | .openclaw/skills/ |
| VS Code + Copilot | — | .github/skills/ |
| Windsurf | ~/.windsurf/skills/ | .windsurf/skills/ |
| Cline | ~/.cline/skills/ | .cline/skills/ |
| OpenCode | ~/.config/opencode/skills/ | .opencode/skills/ |
Official skill: skills/engra-memory/SKILL.md. Copy the entire engra-memory directory to the path for your agent. See the Memory MCP guide for MCP setup.
skills/engra-memory/SKILL.md
More SKILL.md-compatible tools
These tools also support the Agent Skills standard — reuse skills/engra-memory/SKILL.md as-is. MCP setup varies by client; configure the Streamable HTTP endpoint per tool docs: OpenAI Codex CLI (~/.codex/skills/), Gemini CLI, Roo Code, Goose, Amp, JetBrains Junie, and others. See agentskills.io for the full list.
Installation
- Create an API key in the console (/dashboard/keys) with memory:read; add memory:write for saves
- Add MCP config to your client (see example below or skills/engra-memory/mcp.example.json)
- Copy skills/engra-memory/ to your personal or project agent skills directory
- Restart the agent or reload skills; verify memory_list_libraries is available
- On new sessions the agent should follow the skill and call memory_wake_up / memory_search
MCP configuration (skill prerequisite)
The skill teaches how to use memory tools; the tools themselves come from Memory MCP. Configure both: MCP for connectivity, the skill for workflow and trigger timing.
{
"mcpServers": {
"engra-memory": {
"url": "https://mcp.engra.ai/api/v1/memory/mcp",
"headers": {
"Authorization": "Bearer inf_xxxxxxxx",
"Accept": "application/json, text/event-stream",
"Content-Type": "application/json"
}
}
}
}OpenClaw / Claude plugin bundle layout
OpenClaw loads Claude-format bundles natively: skills/ roots become skills, and .mcp.json merges into embedded MCP settings. Package the Engra skill with MCP config for team distribution.
your-team-bundle/ ├── skills/engra-memory/SKILL.md ├── skills/engra-memory/TAXONOMY.md └── .mcp.json
Recommended workflow (in skill)
- memory_list_libraries → obtain memoryLibraryId (use a separate library for sensitive domains)
- New session → memory_wake_up (pass the user's opening message)
- Need prior context → memory_search / memory_recall
- Durable knowledge emerges → proactively memory_save_atom (one fact per atom; self-contained body; search first to dedupe; set scope/topic)
- Outdated search hits → memory_correct_atom (avoid duplicate saves; keep revisions)
- Write tools default to async (202 + job.id): fast ack, background persist; do not expect immediate search hits
- Task done or session ending → save decisions and lessons learned
OpenClaw notes
- Claude commands/ and Cursor .cursor/commands/ map to additional OpenClaw skill roots
- When migrating from Claude, skills with SKILL.md copy into the OpenClaw workspace skills directory
- Imported Claude commands default to disable-model-invocation: true — edit frontmatter for auto-invocation
- See OpenClaw docs: Plugin bundles and Migrating from Claude