Skill: publish a map view of a space as a Claude artifact #199

Goal

A skill in the Claude plugin (plugin/skills/) that publishes a map view of a space as a private Claude artifact, refreshes its data, and reads comments left on the artifact. The renderer, template page and export command live in structured-context-map-views; this skill is the agent-facing steps. Plan and evidence: mindsocket/structured-context-map-views#17.

Steps the skill carries

1. Run structured-context-map-views export <space> (map-views#18) for the graph, validation results, schema id and applicable map view definitions as one JSON document.
2. Fetch the template artifact.html for a pinned map-views version (from the npm package via jsDelivr) and publish it unchanged with the Artifact tool, declaring only the db capability (map-views#19).
3. Write the export into the artifact db with write_db (one spaces/<space> document, JSON as strings). Refreshing a space is the same write with the document's current if_version; the page does not change.
4. Reading comments: a thread's [location] reads like button "<view>/<nodeId>: <type> <title>…" (map-views#20 makes the host's labelling exact). Look the node up in the spaces/<space> document to get its fields, parent and validation messages. The location is the card even when the comment is on a child element of the card; parameter values (orientation, overlay level) are not in it, so ask if they matter.
5. A one-line note that the space's content is stored in the artifact and its database. No confirmation gate.

Decisions already taken

• Placement: a skill in this plugin, not a CLI command in core and not a plugin-bundled dependency. The command lives in map-views (it depends on core's library API; core depending on map-views would be a cycle and would pull React into this CLI). Rationale and rejected options: map-views#17.
• No wait on validate --json: emit the validateSpace result shape, not a second derived one #198: the export calls validateSpace() directly. validate --json: emit the validateSpace result shape, not a second derived one #198 (one validate --json shape) remains worthwhile for consistency.
• Where a setting lives: the skill passes the space name and relies on config as the CLI does (no new environment variables).

Findings the skill should encode

• Comments reach the agent only as [location] (accessible name of the element) and the [anchored at] DOM path. Page-invented anchor names (customAnchors) and onSendToClaude did not work for the viewer tested, so the skill must not suggest them.
• Per-document size limit of the artifact db is unknown; for a large space the skill should chunk (write_db batches of 50) and say so.
• Needs the Artifact tool; not usable from a plain CLI session.

Definition of done (this repo)

• New skill under plugin/skills/; update the command list in plugin/skills/structured-context/SKILL.md if it should be discoverable from there; troubleshooting entries for the failures seen (blank page from a duplicate React, db unavailable, doc missing).
• docs/: one short mention where map views are described, if any; keep detail in the skill.
• Plugin version bumps via the pre-commit hook; do not edit it.

Blocked by

map-views#18 (export), #19 (template and host), and ideally #20 (data-node-id). Write the skill as if these are released, test against local checkouts until they are.