Data Flow

marketplace.json → skill-inventory.js → explorer-data.json → the 3 Explorer views

One Dataset, Three Views

apps/explorer/scripts/generate-explorer-data.js writes exactly one output file, apps/explorer/public/explorer-data.json, which every Explorer view fetches through apps/explorer/src/data.js's cached loadData(). It replaces three legacy per-app generators (skills-registry, dependency-graph, skill-map) whose separate JSON outputs are now just different slices of this one document.

graph LR
  MP["marketplace.json
plugins[] · name/source/skills[]"]
  SKMD["SKILL.md frontmatter
+ git log dates"]

  subgraph LIB["skill-inventory.js"]
    INV["getInventory()
DIVISION_COLORS palette
DIVISION_ALIASES normalize"]
  end

  subgraph GEN["generate-explorer-data.js"]
    MCP["scanMcpServers()
plugins/infrastructure/mcp-servers"]
    EDGES["buildEdges()
skill.dependencies.skills / .mcp"]
    STATIC["DATA_SOURCES · LAB_ITEMS
hand-authored constants"]
  end

  OUT["explorer-data.json
meta · plugins · skills · edges
mcpServers · map"]

  LOADER["data.js
loadData() — fetch + cache"]

  subgraph VIEWS["3 Explorer Views · apps/explorer/src/views/"]
    REG["registry.js
#/registry"]
    MAP["map.js
#/map"]
    GRAPH["graph.js
#/graph"]
  end

  MP --> INV
  SKMD --> INV
  INV --> EDGES
  INV --> MCP
  MCP --> EDGES
  INV --> OUT
  EDGES --> OUT
  MCP --> OUT
  STATIC --> OUT
  OUT --> LOADER
  LOADER --> REG
  LOADER --> MAP
  LOADER --> GRAPH

  class MP,SKMD source;
  class INV,MCP,EDGES,STATIC derive;
  class REG,MAP,GRAPH view;

explorer-data.json — Top-Level Shape

Key	Populated by	Consumed by
`meta`	`getSourceCommitDate()` (only when `SEER_HISTORY_OUTPUT=1`, i.e. the deploy job — `null` in tracked output, see issue #203) + `inventory.divisions`	all 3 views (header stats, division palette)
`plugins`	`inventory.plugins` (from marketplace.json)	`registry.js`
`skills`	`inventory.skills` (SKILL.md frontmatter)	`registry.js`, `map.js`, `graph.js`
`edges`	`buildEdges()` — resolves `metadata.skill-dependencies` / `metadata.mcp-dependencies` by exact id match	`graph.js`
`mcpServers`	`scanMcpServers()` over `plugins/infrastructure/mcp-servers`	`graph.js`
`map.coreSkills`	`inventory.skills` filtered to plugin `core-dependencies`	`map.js`
`map.dataSources` / `map.labItems`	hand-authored `DATA_SOURCES` / `LAB_ITEMS` constants in the generator	`map.js`

Explorer Views ↔ mkdocs Nav

apps/explorer/src/router.js dispatches on a hash route (#/registry, #/map, #/graph), each mapped to a handler registered via registerRoute(). mkdocs.yml's Apps → Explorer nav section links directly to these three hash routes.

Registry — #/registry

Renders plugins + skills as a browsable, filterable list.

Skill Map — #/map

Renders map.coreSkills, map.dataSources, map.labItems plus meta.divisions.

Dependency Graph — #/graph

Renders skills + edges + mcpServers as a node/edge graph.

Silent-drop fix baked into the schema. The legacy skill-map / dependency-graph generators walked hardcoded plugins/divisions/* directories and never looked at plugins/preview or plugins/wil, silently dropping 12 + 17 skills. Because every view now reads the same getInventory()-derived explorer-data.json, that class of drift can't recur per-view — a fix to discovery in skill-inventory.js fixes all three views at once.