<< All versions

Skill v1.0.0

currentAutomated scan100/100
open-science-pillars/core/start
──Details
PublishedSeptember 27, 2026 at 04:22 PM
Content Hashsha256:a1d6bd16a95b409f...
Git SHA1e4b0ecd7533
──Files
Files (1 file, 3.4 KB)
SKILL.md3.4 KBactive
SKILL.md · 70 lines · 3.4 KB

version: "1.0.0" name: start description: "Session orientation: list installed science plugins, connector status, local config, available workflow skills, one suggested next step."


start

Orient the scientist in one screen: what is installed, what is connected, what is configured, and what to do next. Works identically invoked by slash command or by asking conversationally ("what science tools do I have here?").

Behavior

Produce exactly one screen with these five parts, in order:

  1. Installed OSP plugins and their knowledge. Which Open Science

Pillars plugins are present (core; ocean-science and other domain plugins; the provider bundle plugin they depend on, such as nasa-daac-knowledge), with one clause each on what they cover, read from the installer's record (claude plugin list --json where a shell exists; see consult-knowledge for how bundle roots are found from it). One line per plugin: version, the date it was installed or last updated, and its knowledge in numbers: concepts by status (stable, draft, deprecated) and how many stable concepts are past their stale_after date today, counted from the concept frontmatter under each bundle root. A plugin whose record carries errors is reported as disabled with the error's own words. All of this is read from disk; nothing is fetched, and no comparison with the catalog is made. The part closes with one line: installs stay at the release they were installed from, and claude plugin update <name>@open-science-pillars moves one.

  1. Connector status. Whether the Earthdata MCP connector is

configured and reachable this session; when it is not, say so plainly and note that discovery falls back to knowledge-based search (discover-data names the fallback when it fires). On surfaces with per-session connectors, "not configured this session" is a normal state, not an error.

  1. Local config summary. If a project local config exists (for

example ocean-science.local.md), summarize the filled blocks (data paths, compute, region defaults) in a line or two; if none exists, say none is configured and which template creates one.

  1. Available workflow skills, grouped by plugin. Core's workflows

(discover-data, report), then each installed domain plugin's workflows. Knowledge skills are not listed; they load themselves as background expertise.

  1. One suggested next step. Exactly one, chosen from the state

above: a plugin disabled by a dependency error suggests the install command the error names; no data touched yet suggests discover-data; an analysis in progress suggests finishing it and running report; a domain plugin installed but unconfigured suggests filling its local config.

Rules

  • One screen. No scrolling walls, no exhaustive option lists.
  • Read-only: start never loads data, writes files, or asks questions.
  • Never invent status: check what is actually loaded and configured;

when something cannot be checked from this surface (the installer's record on a surface without a shell), say so rather than guessing.

  • Knowledge counts come from the files under the bundle roots, never

from an index line or a remembered figure.

Must NOT

  • Never list knowledge skills in the menu of things to invoke.
  • Never suggest more than one next step.
  • Never trigger a download, a file write, or a connector call that

costs more than a status check.

All versions