memory_self_edit
Shipped in v3-alpha (Beta). Interface may still change before 3.0 stable.
Description
Edit your own memory. Correct errors, strengthen useful knowledge, weaken unreliable facts.
Input schema
{
action: "append" | "replace" | "strengthen" | "weaken" | "reflect",
target: string, // fact_id, "profile", or "strategy:{id}"
content?: string, // for append/replace
reason: string, // MANDATORY — appears in the audit log
}Example
// Strengthen a strategy that just worked
await mcp.callTool("memory_self_edit", {
action: "strengthen",
target: "strategy_abc",
reason: "Used in this incident — restored service in 3min",
});
// Replace an outdated fact
await mcp.callTool("memory_self_edit", {
action: "replace",
target: "fact_xyz",
content: "Redis cluster has 5 shards (was 3 before resize)",
reason: "Confirmed via redis-cli cluster nodes",
});
// Trigger self-reflection cycle
await mcp.callTool("memory_self_edit", {
action: "reflect",
target: "profile",
reason: "End of week — review what worked",
});Constraints
reasonis required — empty reasons are rejected- Rate limits per session (default: 5 edits/session, 3 deletes/day)
- Edits enter a grace period before becoming permanent (default: 300s)
- Some actions require approval per RBAC policy
Backed by
- Self-editing memory — concept + safety model
- Agent RBAC — permission model