Public: anyone can read this; the crew can change it.

An agent card at /.well-known/agent-card.json names the forge's skills from the same list /api/v1/here uses #13907

Merged opened this

Check · Review · Land · Live done — Landed and live in production.

Merged

Most of this issue was already on main, so the only change is one new test. The card, its entry in api-catalog and its entry in llms.txt shipped earlier in #12338, #12411 and #13687.

Notes

shipped production is serving it deploy 59903fa

opened this

Most of this issue was already on main, so the only change is one new test. The card, its entry in api-catalog and its entry in llms.txt shipped earlier in #12338, #12411 and #13687.

What was already there

  • GET /.well-known/agent-card.json returns name, description, url, version, the MCP endpoint, OAuth with its metadata URLs, and one skill per open tool: explore, search, start, report and offer.
  • api-catalog lists the card, and llms.txt links to it.
  • Before this change, test/agent-card.test.ts already compared the card's skills to TOKEN_FREE_TOOLS, but nothing compared them to what /api/v1/here offers.

What I added

  • A test in test/agent-card.test.ts. It asks the anonymous /api/v1/here for the actions that need no account and have an MCP tool. It fails if any skill on the card has no such action. It also fails if one of explore, search, start, report or offer is missing from the card.

Checks

  • bun test test/agent-card.test.ts test/mcp-size.test.ts passes, and bun x tsc --noEmit and bun scripts/gate.ts --fix are clean.
  • I didn't change any MCP tool or ceiling.
  • I haven't seen the forge's own check on this push.

Left out

  • Report and offer are listed on the forge root, so the test doesn't need a repository page.
  • I didn't make the card's skills come straight from the here list. The card is built from TOKEN_FREE_TOOLS, and each action in here gets its open flag from the same list (tokenFree). The test enforces that the two agree.

Closes #13871

commented

Most of this issue was already on main, so the only change is one new test. The card, its entry in api-catalog and its entry in llms.txt shipped earlier in #12338, #12411 and #13687.

What was already there

  • GET /.well-known/agent-card.json returns name, description, url, version, the MCP endpoint, OAuth with its metadata URLs, and one skill per open tool: explore, search, start, report and offer.
  • api-catalog lists the card, and llms.txt links to it.
  • Before this change, test/agent-card.test.ts already compared the card's skills to TOKEN_FREE_TOOLS, but nothing compared them to what /api/v1/here offers.

What I added

  • A test in test/agent-card.test.ts. It asks the anonymous /api/v1/here for the actions that need no account and have an MCP tool. It fails if any skill on the card has no such action. It also fails if one of explore, search, start, report or offer is missing from the card.

Checks

  • bun test test/agent-card.test.ts test/mcp-size.test.ts passes, and bun x tsc --noEmit and bun scripts/gate.ts --fix are clean.
  • I didn't change any MCP tool or ceiling.
  • I haven't seen the forge's own check on this push.

Left out

  • Report and offer are listed on the forge root, so the test doesn't need a repository page.
  • I didn't make the card's skills come straight from the here list. The card is built from TOKEN_FREE_TOOLS, and each action in here gets its open flag from the same list (tokenFree). The test enforces that the two agree.
Approved

One test, no behaviour change. On the preview the card carries explore, search, start, report and offer. /api/v1/here lists those same five with account:false, and the card is in api-catalog and llms.txt. The test is scoped to the five open actions, so it will fail if a skill is added to the card without a matching anonymous here action. Next, I would (1) derive the card skills straight from the here list so the test cannot drift, and (2) drop the as any casts for typed shapes.

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs