Agent Skills
The stock skills teach an agent to use the tools well: when to store, how to choose edge properties, when to correct and when to adjust, and how to orient at the start of a session. They ship as a Claude Code plugin, and their text can be given to any agent.
What the skills are
The tools say what an agent can do; the skills say how to do it well. Each skill is a page of guidance an agent reads when the task calls for it: how to decompose what it learned into records, which edge properties to reuse, how to search and read before relying on what it was trained on, and how to tell a correction from an adjustment.
The engine does not need the skills to run. Without them an agent still has the instructions and tool descriptions the engine hands it over MCP.
The core skills
| Skill | Teaches the agent to |
|---|---|
e-orient | Check its lanes and write schema before storing — so facts land in the right place; survey recent activity, active corrections and project context at session start |
e-store | Split a thought into standalone records and tag them with reusable edge properties |
e-retrieve | Gather → read before reasoning, and trust quartiles over raw distance |
e-correct | Make the wrong/evolved call — correct vs. adjust — instead of duplicating |
e-properties | Reuse the existing vocabulary instead of inventing a new label every time |
e-doctor | Diagnose connection, auth, and profile issues on its own |
e-file-upload | Store a file verbatim, so it can be retrieved later as the same file, possibly by another agent |
e-file-download | Reassemble a stored file from its parts and deliver it back |
Toolkit alone: a database. Toolkit plus skills: an agent that knows how to remember, and a structure that builds itself. Start with the tools to see it work; add the skills to watch it cohere.
The plugin also carries:
- An orientation agent,
e-orient, which surveys memory to restore situational awareness at session start or after the context is compacted. - Session-start hooks, which, when a session starts, resumes or is compacted, remind the agent to load the edge-property vocabulary, its standing directives and its session context from memory.
Installing them
In Claude Code, install the skills as a plugin from the eMemory repository:
claude plugin marketplace add everythingmemory/home
claude plugin install ememory@everythingmemory
Or, inside Claude Code, run /plugin and pick ememory from the everythingmemory marketplace. Installing eMemory never installs the plugin; it is your choice.
Then connect the agent to the API service; see connecting clients.
Other agents and harnesses
Every agent that connects over MCP receives the engine's instructions and the description of each tool, whatever its harness. For a harness without plugins, give the agent the skills' text as part of its instructions.
Licence
The skills, the orientation agent and the hooks are licensed under the Apache License 2.0. See licence.