A fork's page says in one line how it follows its upstream and an admin switches it in one tap #10765

A fork's page now says in one line how it follows its upstream, and an admin switches the mode with one button after a confirm. I ran the gate, tsc, design-lint and the seven affected test files, including test/mcp-size.test.ts. They passed, 60 tests, 0 failures. I did not run the full suite.

What changed

• The line: the Following card on a fork's repository page shows Follows <upstream> automatically or Proposes updates from <upstream>, with the reason beside it. Anyone who can see the card sees the line. followModeWords and followSwitchPreview in src/actions/follow_words.ts hold the wording.
• The switch: for an admin who can write, one button (“Follow automatically” or “Only propose updates”) replaces the old radio form. It goes to /-/confirm, which previews what changes and, for automatic, says a conflict still goes to the crew. Confirming POSTs to the new route /:owner/:name/following/mode/:mode. That POST goes through readForm, so it has the same-origin check, and setFollowMode still receipts it as a forge.follow_mode event.
• MCP: the existing follow_mode tool already switched the mode, so I added no action to repo and no tool. It now answers in the same words as the page (followModeSaid). The tool list is unchanged, so test/mcp-size.test.ts passes without touching the ceilings.
• Conflicts: automatic mode still runs the existing takeUpdates path, and when both sides changed the same files the fork is told it diverged so the crew converges it. I did not touch that path.
• Tests: test/fork-tree-web.test.ts covers both modes on the page, the confirm preview, the cross-origin POST (403), a non-admin POST (303, nothing saved) and the audit event. A reader sees the line but no switch. test/spawn-chain-mode.test.ts checks the MCP words.

Things to know

• Reserved query key: ?mode= is reserved by the server (the old studio/dev redirect strips it), so the mode travels in the path rather than the query.
• Replaced test: I replaced the old radio-form assertions in test/fork-tree-web.test.ts, because that UI no longer exists on the fork page. The forge's own chain page keeps its radio form.
• Not changed: the TUI toasts still say “Now following in X mode.” I left them alone.
• Empty commit: the pre-push hook wanted a “Try:” commit, so the branch ends with an empty one.

To try it: open a fork you administer, read the line, and press the button. Confirm, and the line flips.

🤖 Generated with Claude Code

Closes #10748

A fork's page now says in one line how it follows its upstream, and an admin switches the mode with one button after a confirm. I ran the gate, tsc, design-lint and the seven affected test files, including test/mcp-size.test.ts. They passed, 60 tests, 0 failures. I did not run the full suite.

What changed

• The line: the Following card on a fork's repository page shows Follows <upstream> automatically or Proposes updates from <upstream>, with the reason beside it. Anyone who can see the card sees the line. followModeWords and followSwitchPreview in src/actions/follow_words.ts hold the wording.
• The switch: for an admin who can write, one button (“Follow automatically” or “Only propose updates”) replaces the old radio form. It goes to /-/confirm, which previews what changes and, for automatic, says a conflict still goes to the crew. Confirming POSTs to the new route /:owner/:name/following/mode/:mode. That POST goes through readForm, so it has the same-origin check, and setFollowMode still receipts it as a forge.follow_mode event.
• MCP: the existing follow_mode tool already switched the mode, so I added no action to repo and no tool. It now answers in the same words as the page (followModeSaid). The tool list is unchanged, so test/mcp-size.test.ts passes without touching the ceilings.
• Conflicts: automatic mode still runs the existing takeUpdates path, and when both sides changed the same files the fork is told it diverged so the crew converges it. I did not touch that path.
• Tests: test/fork-tree-web.test.ts covers both modes on the page, the confirm preview, the cross-origin POST (403), a non-admin POST (303, nothing saved) and the audit event. A reader sees the line but no switch. test/spawn-chain-mode.test.ts checks the MCP words.

Things to know

• Reserved query key: ?mode= is reserved by the server (the old studio/dev redirect strips it), so the mode travels in the path rather than the query.
• Replaced test: I replaced the old radio-form assertions in test/fork-tree-web.test.ts, because that UI no longer exists on the fork page. The forge's own chain page keeps its radio form.
• Not changed: the TUI toasts still say “Now following in X mode.” I left them alone.
• Empty commit: the pre-push hook wanted a “Try:” commit, so the branch ends with an empty one.

To try it: open a fork you administer, read the line, and press the button. Confirm, and the line flips.

🤖 Generated with Claude Code

All three blockers and the inline comments are addressed and pushed. tsc, the gate and the design lint are clean. The six test files I ran pass, along with test/tui-following.test.ts; I did not run the full suite.

What changed

• Both doors.

- A fork's terminal screen now shows the mode line ("Proposes updates from ann/yard" or "Follows ann/yard automatically") instead of "tells you" / "takes updates". - Pressing m toasts followModeSaid(...), the same sentence the web page and MCP use. - The forge-chain terminal screen (src/tui/screens/chain.ts) also dropped "Now following in X mode" for followModeSaid. - I updated the existing tests that asserted the old wording: test/spawn-chain-doors.test.ts and test/fork-follow-mode.test.ts.

• Docs. New docs/features/follow-mode.md covers the line, the confirm, the MCP follow_mode tool and the m key. README is untouched.
• Tests.

- The audit is asserted as exactly one more event, with this repo, this actor and the mode. - The MCP test now covers the "Proposes updates from" wording, a non-admin call being refused, and the switch back. - New web test: the tell-me confirm preview, the "its parent" preview for a non-writer, a root repository with no line, and followModeSaid(undefined, …) with no parent name.

• Cleanups.

- In src/web/pages/search.ts the unreachable "Pick a mode" branch is gone; the route pattern already limits the mode, so it is cast to FollowMode. - The Following card says the mode once, as the line. The reason sentence stays in followModeSaid for MCP and the terminal toast, and in the confirm preview. - FollowMode from spawn/chain.ts is the only definition of the mode type in follow_words.ts.

Not done: the confirm preview for a non-writer still says "its parent". You called that fine, so I only added the test.

Try it: open a fork's page as its admin and tap "Follow automatically". Or open the fork's screen over SSH and press m.

🤖 Generated with Claude Code

I couldn't reproduce either failure here. I merged the latest main into the branch and pushed it, so I can't say what caused them.

On the old head, both tests (docs/MCP.md holds the tool table… and the page offers the three windows…) passed when run alone and in a repeat loop. bun scripts/mcp-table.ts made no change to docs/MCP.md, so the tool table was already in sync. I believe the check ran against a different main than my branch was based on, and the merge fixes that. I haven't confirmed it.

After the merge, bun x tsc --noEmit is clean and test/mcp-table.test.ts and test/traffic-history.test.ts pass. I didn't run scripts/gate.ts or any other test files.

If the check fails again on the same two tests, send me the log lines that name what differed, since I have nothing else to go on.

The pull itself is unchanged: a fork's page says in one line how it follows its upstream, with a confirmed one-tap switch. The terminal and MCP follow_mode use the same words.

🤖 Generated with Claude Code