菜单
Read, search, and edit the user's xNet workspace (pages, databases, canvases) via vault files and the xnet CLI.
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。
Write a markdown exploration document for a topic, feature idea, architecture question, or research area. Use when asked to "explore X", "research X", "write an exploration", "investigate X", or "do a deep dive on X". Produces a numbered doc in docs/explorations/ with mermaid diagrams, callouts, collapsible sections, status tables, recommendations, and checklists.
Autonomously implement what was just written — an /explore exploration, a plan, or a spec in the conversation. Use when the user says "implement it", "build it", "ship it", "do it", "make it so", "implement this exploration/plan", or "this looks good, go". Works on a new branch, checks off checklist items and commits as it goes, opens a PR, and merges to main.
Write a changeset for the current change. Use when you've edited a publishable packages/* library and need to record a semver bump + consumer-facing release note before the turn ends (the Stop hook enforces this). Reads the diff, picks the bump per affected package, and writes .changeset/*.md.
Use for any question about a codebase, its architecture, file relationships, or project content — especially when graphify-out/ exists, where the question should be treated as a graphify query first. Turns any input (code, docs, papers, images, videos) into a persistent knowledge graph with god nodes, community detection, and query/path/explain tools.
| name | xnet |
| description | Read, search, and edit the user's xNet workspace (pages, databases, canvases) via vault files and the xnet CLI. |
This folder is a checkout of the user's xNet workspace. The xNet database stays the source of truth; these files are a working tree. Identity lives in frontmatter and the manifest, never in filenames.
xnet search/query/db get print plain stdout; grep and
pipe them. Reach here first: it costs the fewest tokens.xnet commit.xnet MCP tools are a fallback for shell-less clients; when
the CLI can do it, the CLI is cheaper.Reads are free; writes are deliberate. Edits become validated mutation
plans that only persist on xnet commit --apply; the MCP server is read-only
unless it was connected with --writes. Scope every checkout (--query,
--schema, --node, --kind) — never materialize the whole workspace.
Pages/<slug>.md — pages as Markdown with YAML frontmatter
(xnet.id, xnet.revision). Never edit frontmatter.Databases/<slug>.schema.json + <slug>.rows.jsonl — one JSON object
per line; edit, append, or remove lines to change rows.
<slug>.tsv sidecars are read-only fast reads.Canvases/<slug>.canvas — JSON Canvas projections.[[Title]], :::xnet-database blocks, and
{{xnet-ref ...}} directives are live references — preserve them.xnet search "<text>" for ranked
results (TSV: id, slug, title, snippet).xnet checkout --query "<text>" (or --schema <iri>,
--kind page|database|canvas, --node <id>) materializes a scoped
slice into this folder. Never export everything.xnet query <database> --where field=value --format tsv
(default TSV; --format jsonl|json when you need structure).xnet daemon running
they are picked up automatically; otherwise run xnet commit.xnet status lists pending plans and conflicts. Conflicts are quarantined
in .xnet/conflicts/ with a Markdown note explaining how to resolve;
fix the file and save again. Stale edits never silently overwrite.xnet run <script.js> executes a sandboxed
script with an api object (api.nodes(schema), api.search(text),
api.proposeUpdate(id, props), api.proposeCreate(schema, props)).
Writes are proposals that flow through the same plan pipeline. Return a
digest, not raw rows.