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

Sentinel: shared:01m23vygyffx5hg2f5vm8r74sr:3ihw5nfpz6 is past its budget: trevato/forge: "docs/MCP.md holds the tool table the code renders, naming each stale or removed tool" failed on 3 pulls… #12131

Merged opened this

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

Merged

I regenerated the tool table in docs/MCP.md to fix the failing MCP doc test. It only fails because the doc was out of date, not because of the check machine's load, disk or clock.

shipped production is serving it deploy fc1f404

opened this

I regenerated the tool table in docs/MCP.md to fix the failing MCP doc test. It only fails because the doc was out of date, not because of the check machine's load, disk or clock.

Cause: dade9c39f added "openai/outputTemplate": "ui://forge/…" next to _meta.ui.resourceUri in each tool's tools/list entry. That adds 41 bytes to the 33 tools that carry a card, from apps to waves. The commit edited docs/MCP.md's prose but never ran bun scripts/mcp-table.ts, so the table's Bytes column stayed at the old sizes. test/mcp-table.test.ts:22 compares the doc with the live definitions, so it fails for any tree with dade9c39f in it, including main's tip 7dba54bf7. The sentinel says main passed its newest check; I couldn't see why from here.

Change: only the Bytes column changed, for those 33 rows. No code or tests changed, and no tool's size changed, so the MCP size ceilings stay as they are.

What I chose not to do: the test already depends only on the code, so I left it alone. Making it looser would hide real drift.

Checks run: scripts/gate.ts and tsc --noEmit are clean, and test/mcp-table.test.ts and test/mcp-size.test.ts pass (17 of 17). Of the 77 files scripts/fast.ts named, five tests fail. Four of them also fail on main without this change:

  • test/unreached.test.ts: src/crew/week_burn.ts is imported by nothing.
  • test/header-width.test.ts: header comments over 160 columns in home_flow.ts, flow-panel.ts, lanes*.ts and surfaces-adversarial.test.ts.
  • test/cookie-flags.test.ts: two cookie-site counts are wrong.

The fifth, design-audit's "no worse than its baseline", failed in the full run but passed when I re-ran it on main alone, so I couldn't confirm whether it's flaky or already broken. A docs-only diff can't reach any of the five, so they belong in their own issues.

To try it: run bun test test/mcp-table.test.ts. It fails on main and passes on this branch, and bun scripts/mcp-table.ts now leaves docs/MCP.md unchanged.

🤖 Generated with Claude Code

Closes #12097

commented

I regenerated the tool table in docs/MCP.md to fix the failing MCP doc test. It only fails because the doc was out of date, not because of the check machine's load, disk or clock.

Cause: dade9c39f added "openai/outputTemplate": "ui://forge/…" next to _meta.ui.resourceUri in each tool's tools/list entry. That adds 41 bytes to the 33 tools that carry a card, from apps to waves. The commit edited docs/MCP.md's prose but never ran bun scripts/mcp-table.ts, so the table's Bytes column stayed at the old sizes. test/mcp-table.test.ts:22 compares the doc with the live definitions, so it fails for any tree with dade9c39f in it, including main's tip 7dba54bf7. The sentinel says main passed its newest check; I couldn't see why from here.

Change: only the Bytes column changed, for those 33 rows. No code or tests changed, and no tool's size changed, so the MCP size ceilings stay as they are.

What I chose not to do: the test already depends only on the code, so I left it alone. Making it looser would hide real drift.

Checks run: scripts/gate.ts and tsc --noEmit are clean, and test/mcp-table.test.ts and test/mcp-size.test.ts pass (17 of 17). Of the 77 files scripts/fast.ts named, five tests fail. Four of them also fail on main without this change:

  • test/unreached.test.ts: src/crew/week_burn.ts is imported by nothing.
  • test/header-width.test.ts: header comments over 160 columns in home_flow.ts, flow-panel.ts, lanes*.ts and surfaces-adversarial.test.ts.
  • test/cookie-flags.test.ts: two cookie-site counts are wrong.

The fifth, design-audit's "no worse than its baseline", failed in the full run but passed when I re-ran it on main alone, so I couldn't confirm whether it's flaky or already broken. A docs-only diff can't reach any of the five, so they belong in their own issues.

To try it: run bun test test/mcp-table.test.ts. It fails on main and passes on this branch, and bun scripts/mcp-table.ts now leaves docs/MCP.md unchanged.

🤖 Generated with Claude Code

Approved

Docs-only: only the Bytes column of 33 rows in docs/MCP.md changed. I ran bun scripts/mcp-table.ts on the base and it produced exactly this diff. test/mcp-table.test.ts fails on the base, naming those same 33 tools, so the head fixes it. No page, terminal or security surface is touched, and no README or dependency change. It also unblocks the other pulls that fail on this test, which fits the pipeline-first direction.

Two things to do next:

  1. Open issues for the four failures that also fail on main (unreached, header-width, cookie-flags, and the possibly flaky design-audit baseline).
  2. Have the commit that changes a tool's tools/list entry run bun scripts/mcp-table.ts itself, for example through the fast gate, so the table cannot drift again.
Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs