J Gateway Interaction

J is a memory-first AI agent with a durable vault of knowledge, append-only conversation threads, and a governance layer for memory changes. It runs as a local daemon with a WebSocket gateway. This skill lets Claude Code communicate with J — to delegate tasks, retrieve knowledge from its long-term memory, or check on background work.

Why Talk to J?

J maintains persistent, auditable memory across conversations — knowledge about people, projects, preferences, and past decisions. Unlike Claude Code's ephemeral context, J's vault persists between sessions and is searchable via embeddings. Use J when you need:

• •Long-term context: "What did Jesse decide about the auth approach last week?"
• •Background research: Delegate a deep-think task and check back later
• •Knowledge retrieval: Query J's vault for information it has accumulated
• •Task delegation: Hand off work that benefits from J's tools (knowledge search, image generation, deep thinking)

J and Claude Code are complementary: Claude Code is fast and has filesystem access; J has durable memory and domain-specific tools.

When to Use

Use this skill when the user wants to:

• •Ask J a question or send it a task ("ask J about...", "send to J", "talk to J")
• •List or manage J sessions ("J sessions", "show my J conversations")
• •Check what J said or read conversation history ("check J", "what did J say")
• •Retrieve something from J's memory or knowledge vault

Prerequisites

The J gateway daemon must be running. If commands fail with connection errors, suggest:

j gateway start

The binary is j. The daemon binds to 127.0.0.1:9123.

Commands

List sessions

j gateway list

Returns a JSON array of sessions with session_key, thread_id, created_at, title, first_user_line.

Open/create a session

j gateway open <session_key>
# Default session key is "main"
j gateway open

Returns session metadata as JSON. Creates the session (and backing thread JSONL) if it doesn't exist.

Read session history

j gateway history <session_key> --limit 20

Returns {"events": [...], "count": N}. Default limit is 50, max 500. Events include user_message, assistant_message, tool_call, tool_result, system_note, inner_monologue, title_generated.

Send a message (fire-and-forget)

j gateway send <session_key> "your message here"

Returns {"status": "accepted"} immediately. The agent processes in the background. Use for long-running tasks you don't need to wait on.

Send a message and wait for response

j gateway send <session_key> "your question here" --wait
# With custom timeout (seconds):
j gateway send <session_key> "your question here" --wait 60

Streams JSON events to stdout (one per line): tool_call_start, tool_call_result, final, error, deep_think_complete, title_generated. Exits after final or error event. Default timeout: 120s.

Check daemon status

j gateway status

Usage Guidelines

• •Use --wait when querying J for information (questions, lookups, context retrieval)
• •Use fire-and-forget when delegating tasks (long research, background work) — check back with history later
• •Default session is "main" — use it unless the user specifies otherwise
• •Parse JSON output and present results as readable markdown to the user
• •If you get a session.busy error, the agent is already processing — wait and retry, or check history later
• •All output is JSON — parse it, don't show raw JSON to the user unless they ask
• •When reading history, look for assistant_message events for J's responses and tool_call/tool_result for what tools it used

Event Types in History & Streaming

Error Handling

Errors are printed to stderr as JSON with error and message fields:

• •daemon_not_running — suggest j gateway start
• •auth_failed — token issue, suggest restarting daemon
• •session_busy — agent already running on this session, wait and retry
• •timeout — agent took too long, check history later
• •connection_lost — daemon crashed or network issue