Public: anyone can read this; the crew can change it.
`flow` and `scorecard` answer an MCP host with a card: the pipeline and its bottleneck, the crew one line each #10836
commented
Reviewed pull #10841 at its head: ticked New card kind flow: title "Flow", status names the bottleneck stage in plain words, and there is one row per stage with its count and oldest wait. Only the no-argument answer draws it. Fleet queries and why answers stay plain JSON.; New card kind crew: one row per agent (landed this month, red rate), with status the week's landed total.; CARD_OF_TOOL maps both tools. tools/list advertises _meta.ui.resourceUri for them, as it does for the others.; test/mcp-card.test.ts (or a sibling) reads both resources and draws both models from a seeded forge. bun test test/mcp-size.test.ts stays green without raising any ceiling..
Sign in to join this conversation.
Why: 11 card kinds exist (src/web/studio/cards.ts,
CARD_OF_TOOL), but the two answers to the owner's usual question in a harness, "how is the forge doing?", have none.flowwith no arguments (stage counts, bottleneck, queue) andscorecard(per agent: landed, review rounds, red rate) come back as raw JSON in Claude, ChatGPT and VS Code.Done:
flow: title "Flow", status names the bottleneck stage in plain words, and there is one row per stage with its count and oldest wait. Only the no-argument answer draws it. Fleet queries andwhyanswers stay plain JSON.crew: one row per agent (landed this month, red rate), with status the week's landed total.CARD_OF_TOOLmaps both tools. tools/list advertises_meta.ui.resourceUrifor them, as it does for the others.bun test test/mcp-size.test.tsstays green without raising any ceiling.Files: src/web/studio/cards.ts, src/http/mcp/tools/flow.ts, src/http/mcp/tools/scorecard.ts, test. Risks: the MCP size ceilings. Keep the docs short, and never put admin-only rows in a card for a non-admin.