Local-first 🧠memory for Hermes Agent that lives in your Obsidian vault and remembers project context. Nightly 😴 dream passes turn repeat corrections into confirmed preferences with measurable confidence. Adapters ship for Claude Code, Codex, and OpenClaw, with an MCP server for anything else.
TypeScript
/**
* Hook event names allowed to carry
* `hookSpecificOutput.additionalContext` back to the runtime.
*
* Default-closed: any event not listed here gets NO stdout from
* context-injecting hooks. Claude Code validates hook output against a
* per-event schema and rejects `additionalContext` for events that do
* not support it - the rejection both drops the payload and echoes it
* into the validation error, so emitting on an unsupported event is
* strictly worse than staying silent.
*
* `PostCompact` is deliberately absent: current Claude Code has no
* such hook event at all (verified against
* https://code.claude.com/docs/en/hooks, 2026-06-02). Post-compaction
* re-injection is served by the `SessionStart` hook with the
* `compact` matcher instead - see `hooks/hooks.json`.
*
* `SubagentStop` is deliberately absent for the opposite reason: the host
* DOES accept `additionalContext` there, and `hooks/hooks.json` registers
* the event, but only for capture - no hook in this tree injects context
* when a delegated sub-agent stops. Listing it would be a permission with
* no producer behind it, and this list is a statement about what is
* emitted, not about what would be accepted.
*/
export const CONTEXT_EVENT_NAMES = Object.freeze(["SessionStart", "UserPromptSubmit"] as const);
export type ContextEventName = (typeof CONTEXT_EVENT_NAMES)[number];
export function isContextEventName(name: string): name is ContextEventName {
return (CONTEXT_EVENT_NAMES as ReadonlyArray<string>).includes(name);
}