菜单
Check that Neuralscape is reachable and show the current plugin configuration. Use when the user asks "is neuralscape working", "is my memory service up", "what URL is neuralscape using", or any health/connectivity question. Works in both Claude Code and Claude Cowork.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Ingest files, folders, or a block of pasted context into Neuralscape memory. Use when the user says "ingest this file/folder", "add these docs to memory", "index this PDF/Word doc", "load this documentation into Neuralscape", or pastes a long passage they want remembered. Handles Markdown, HTML, PDF, and MS Office (docx/xlsx/pptx); files are stored as artifacts and referenced back. Works in both Claude Code and Claude Cowork.
Load the user's relevant Neuralscape memories before planning or acting — their preferences, conventions, tech stack, and past decisions. Use at the start of a task, when the user says "neuralscape on" / "load my context" / "what do you know about me or this project", or whenever you need prior context. Works in both Claude Code and Claude Cowork (MCP-driven).
Extract and save the durable facts from the current conversation to Neuralscape at the end of a substantive working session. Use when the user says "save this session", "save this conversation", "compile this session", or when wrapping up meaningful work. Works in both Claude Code and Claude Cowork (MCP-driven) — and is the Cowork stand-in for the Claude Code Stop hook.
Search the user's Neuralscape memory store for relevant facts. Use when the user asks "what do I know about X", "have we discussed Y", "what did I decide about Z", or any retrieval-style question. Returns hybrid vector + graph results. Works in both Claude Code and Claude Cowork (MCP-driven).
Save one durable fact to Neuralscape memory now — a preference, decision, convention, gotcha, or architectural choice. Use when the user says "remember this", "save that", "note that I prefer X", or when you learn something worth keeping for future sessions. Works in both Claude Code and Claude Cowork (MCP-driven).
Read a JSONL buffer of captured tool observations and extract significant memories using the Neuralscape memory model v2. Use this when a UserPromptSubmit, SessionStart, or `/neuralscape:capture` instruction asks you to compile observations from a buffer file path.
| name | ns-status |
| description | Check that Neuralscape is reachable and show the current plugin configuration. Use when the user asks "is neuralscape working", "is my memory service up", "what URL is neuralscape using", or any health/connectivity question. Works in both Claude Code and Claude Cowork. |
Confirm Neuralscape is reachable and show what the plugin is currently using. The reachability check uses the MCP connector (works in both platforms); the local /health probe is an additional Claude Code readout when a service URL is configured.
MCP reachability (primary, both platforms) — call list_memories(limit: 1) via the MCP tool. If it returns without error, the connector is reachable; report MCP connector: reachable. If it errors, report the error verbatim and that the connector is unreachable.
Background queues (MCP, both platforms) — call queue_status(). Report the per-queue pending depths (main/graph/ingest) and the caught_up flag — this answers "are my writes done?" without any REST polling.
Read config (Claude Code) — URL from process.env.CLAUDE_PLUGIN_OPTION_URL (fallback NEURALSCAPE_URL); user_id from CLAUDE_PLUGIN_OPTION_USER_ID (fallback NEURALSCAPE_USER_ID); API-key state only from CLAUDE_PLUGIN_OPTION_API_KEY (fallback NEURALSCAPE_API_KEY) — never echo the value.
Local /health probe (Claude Code only, when URL + curl present) — GET <URL>/health with a 5-second timeout, Authorization: Bearer <key> if set. Report the per-backend checks. This is the one sanctioned REST call in this skill — an ops probe of the backing stores that has no MCP equivalent. (/health/live is the dependency-free liveness endpoint; /health is the deep readiness check that probes Redis/Qdrant/Neo4j.) Never use REST for any memory operation.
Render a compact status block. Two shapes depending on what's available:
Claude Code (env + URL present):
Neuralscape — status
MCP connector: reachable
queues: main 0 · graph 1 · ingest 0 (caught up: no)
URL: https://neuralscape.example.com
user_id: aydin
API key: set
/health: 200 OK
redis: ok
vector_store: ok
graph_store: ok
Cowork / connector mode (no local env/URL):
Neuralscape — status
MCP connector: reachable
queues: main 0 · graph 0 · ingest 0 (caught up: yes)
identity: from OAuth token (no local user_id/URL set — expected in Cowork)
/health: not available here (no local service URL)
If the local /health call fails (timeout, 5xx, connection refused) in Claude Code, report the error verbatim and suggest checking docs/neuralscape/01-getting-started.md Step 4 in the Neuralscape repo. Do not treat a missing local URL as an error in Cowork — connector mode is normal there.
ns-config skill — in Claude Code those live in the keychain (sensitive) or settings.json (non-sensitive) via /plugin config neuralscape@neuralscape-plugins; in Cowork identity comes from the OAuth connector, not local config.