| name | echovault |
| description | Local-first memory for coding agents. You MUST retrieve task-aware context before substantive work and save durable learnings before session end. |
EchoVault — Agent Memory System
You have persistent memory across sessions. USE IT.
Before feature, planning, debugging, or architecture work — MANDATORY
Use the current user request as the query so retrieval is task-aware. Prefer the
memory_context MCP tool:
{
"project": "<current-project>",
"agent": "claude-code",
"query": "<current user request>",
"token_budget": 1200
}
CLI fallback:
memory context --project --agent claude-code --query "<current user request>"
If the context pack is insufficient, use memory_search for a narrower topic.
Fetch memory_details only when the returned summaries and structured context are
not enough. If context policy reports disabled, continue normally; explicit
search and save remain available.
Do not replace the current request with generic terms such as "recent context".
Session end — MANDATORY
Before ending your response to ANY task that involved making changes, debugging, deciding, or learning something, you MUST save a memory. This is not optional. If you did meaningful work, save it.
memory save \
--title "Short descriptive title" \
--what "What happened or was decided" \
--why "Reasoning behind it" \
--impact "What changed as a result" \
--tags "tag1,tag2,tag3" \
--category "<category>" \
--related-files "path/to/file1,path/to/file2" \
--source "claude-code" \
--details "Context:
Options considered:
- Option A
- Option B
Decision:
Tradeoffs:
Follow-up:"
Categories: decision, bug, pattern, learning, context, playbook,
known_fix, constraint, project_state, active_work.
Use --source claude-code to identify the agent.
What to save
You MUST save when any of these happen:
- You made an architectural or design decision
- You fixed a bug (include root cause and solution)
- You discovered a non-obvious pattern or gotcha
- You set up infrastructure, tooling, or configuration
- You chose one approach over alternatives
- You learned something about the codebase that isn't in the code
- The user corrected you or clarified a requirement
What NOT to save
- Trivial changes (typo fixes, formatting)
- Information that's already obvious from reading the code
- Duplicate of an existing memory (search first)
Agent setup (recommended)
Run once to install MCP and refreshed behavioral instructions for your agent:
memory setup claude-code
To remove: memory uninstall claude-code
Other commands
memory config
memory sessions
memory reindex
memory delete <id>
Rules
- Retrieve before working. Save before finishing. No exceptions.
- Always capture thorough details — write for a future agent with no context.
- Never include API keys, secrets, or credentials.
- Wrap sensitive values in
<redacted> tags.
- Search before saving to avoid duplicates.
- One memory per distinct decision or event. Don't bundle unrelated things.