A hot-file hold names the open pulls on that file, oldest first, on the map and in the Flow panel #7491

A hot-file hold now names the open pulls on that file, oldest first, at most 8. They show in the map answer and as links in the Flow panel's "Ready, and why it waits" card. tsc is clean and the new test passes with map-held, boards-held, crew-held and crew-hot-files. I did not run the full suite.

• Where the numbers come from: a new pullsTouching(repoId, path, admit) in src/crew/hot_files.ts reads the per-pull cache the picker already fills, so there is no new git call. Wide pulls are skipped, as they already are in the heat counts. Sealed pulls are skipped for viewers who may not see them, as the sentence's count already does. Oldest first means lowest pull number first.
• Map: mapHolds in src/crew/held.ts adds pulls to a hot_file hold. GET …/map returns it on held for each held issue. The detail sentence is unchanged. Holds set by an admin pin, or waiting on a relieving issue, carry no pulls.
• Flow panel: the issue named the panel file flow-panel.ts, but the hold sentences live in the held card, so I put it there. src/flow/held.ts gives each row up to 5 hot-file holds, and src/web/pages/boards-held.ts renders #N waits: <detail> followed by links #a #b … to each pull. The card is the same one the Flow panel and /boards draw. The new elements are plain ul/li, so design-lint is clean.
• Test: test/held-hot-pulls.test.ts seeds pulls out of order and checks the numbers and their order in the map answer and in the card HTML. It also checks the cap of 8, and that a pull on another file is not named.

I did not add the pulls to the feed line (hotSays), the terminal Flow dialog, or the MCP map tool's own text.

Closes #7423

A hot-file hold now names the open pulls on that file, oldest first, at most 8. They show in the map answer and as links in the Flow panel's "Ready, and why it waits" card. tsc is clean and the new test passes with map-held, boards-held, crew-held and crew-hot-files. I did not run the full suite.

• Where the numbers come from: a new pullsTouching(repoId, path, admit) in src/crew/hot_files.ts reads the per-pull cache the picker already fills, so there is no new git call. Wide pulls are skipped, as they already are in the heat counts. Sealed pulls are skipped for viewers who may not see them, as the sentence's count already does. Oldest first means lowest pull number first.
• Map: mapHolds in src/crew/held.ts adds pulls to a hot_file hold. GET …/map returns it on held for each held issue. The detail sentence is unchanged. Holds set by an admin pin, or waiting on a relieving issue, carry no pulls.
• Flow panel: the issue named the panel file flow-panel.ts, but the hold sentences live in the held card, so I put it there. src/flow/held.ts gives each row up to 5 hot-file holds, and src/web/pages/boards-held.ts renders #N waits: <detail> followed by links #a #b … to each pull. The card is the same one the Flow panel and /boards draw. The new elements are plain ul/li, so design-lint is clean.
• Test: test/held-hot-pulls.test.ts seeds pulls out of order and checks the numbers and their order in the map answer and in the card HTML. It also checks the cap of 8, and that a pull on another file is not named.

I did not add the pulls to the feed line (hotSays), the terminal Flow dialog, or the MCP map tool's own text.

I've fixed the sealed-pull leak and pushed the rest of the feedback to the same branch. I didn't change the MCP map tool's description text. tsc is clean and the new, held, flow and TUI tests I ran pass. I did not run the full suite.

• Leak: readHeld in src/flow/held.ts now caches only each repository's Held, which is the same for every reader. It builds the rows on each call with mapHolds(repo, held, admits(...)). The owner's call counts sealed pulls; a guest's call doesn't. The sentence's count and the listed pull numbers come from the same filter. admits is now exported from src/crew/held.ts. A guest whose file drops under the hot threshold once the sealed pull is excluded gets no hold at all.
• Guest test: the new test seeds pull 26 as sealed on src/hot.ts plus an unsealed pull 27. The guest's card and terminal lines read "in 5 open pulls" with no /pulls/26 or #26. The owner's card reads "in 6 open pulls" and links /pulls/26.
• Terminal: heldLines in src/tui/screens/flow.ts now adds a line per hot hold, #N waits: <detail>: #a #b …. It uses the same rows and the same reader gate as the page, and a test covers it.
• README and MCP:

- The README now describes the pulls on all three doors. - The MCP map tool's JSON already carries held.pulls. - I tried adding a mention to the tool description, but that pushed the tool list over test/mcp-size.test.ts's 38,500-byte limit, so I reverted it. If you want the tool text itself to name the pulls, something else has to give 40 bytes or so.

• Clean: the h.pulls! assertion is gone, replaced by a single flatMap. The test header now says #7491. The test shares one SEED constant instead of re-seeding by hand.
• Links: the pulls sit in a .held-pulls wrapping flex line, and each link is 44px tall with 8px of padding. The class is named in docs/design/areas/boards.md, and design-lint is clean.

The full gate on land was red; the pull is open again. Red tests:

• 200 branches hold the 50 ms budget and name their phases
• the three planned bodies are held by nothing on their own hot files; a plain one on the same file is

The gate's log: /api/v1/repos/trevato/forge/gates/1793141610761-2165d743e0e7/log

A hot-file hold on the map and in the held card now names the open pulls on that file, oldest first. It does this for the API, MCP and terminal Flow dialog, and for each reader separately, so sealed pulls don't leak. The gate's second failure is fixed. The first, the branches timing test, I did not change, and it may fail again on the gate.

• held-relieves-map test: test/held-relieves-map.test.ts came in with main (#7496) and was written before holds carried pulls. It compared the plain issue's held object exactly, so the new pulls field broke it. The test was out of date, not the code. It now matches by and detail and checks that pulls lists all HOT_PULLS + 1 pulls on the file. I merged main into this branch to bring the test in.
• "200 branches hold the 50 ms budget": this passes here (test/branches-many.test.ts, 7 of 7), and nothing in this pull touches the branches page. The gate timed it at 276 ms on a machine running 4 shards. I read that as load noise and made no change.

bun x tsc --noEmit is clean. test/held-relieves-map.test.ts, test/hot-files-relieves.test.ts and test/branches-many.test.ts pass.

I left out juniper's two non-blocking items:

• Naming the pulls in the MCP map tool description waits on the tools/list budget (#6217).
• Caching the per-reader result for the 5 s TTL only matters if the boards page gets slow.

I changed nothing for this failure, so there is nothing new to push.

The failing test, GET /organizations/new names its phases and answers within its budget, is in test/cloud-org-apps-timing.test.ts. It passes here, and the whole file passes: 5 of 5 in 1 s.

I don't think my change can cause it. The diff against main touches only the hot-file hold code (src/crew/held.ts, src/crew/hot_files.ts, src/flow/held.ts), the terminal Flow screen, the held-files card on /boards, one CSS component block, the README, a design doc and tests. /organizations/new doesn't use any of those. The test's own comment says timers jitter on a busy machine, and the check ran it in a 4-shard parallel run, where it failed at 56 ms.

If the check turns red on this test again with no code change, the test's margin needs widening in a separate pull. I'd rather not loosen an unrelated budget test inside this one.

The pull's description from my last message still stands: hot-file holds name the open pulls, resolved per reader, on the map, in the held card and in the terminal Flow dialog.

I merged main into the branch and pushed it (cd69261f). The only conflict was in src/flow/held.ts.

• Resolution: held rows keep both sides' changes. main's small field (the small-issue share from #7571) stays, and so does this pull's hot list of hot-file holds with the open pulls on each file. The cached entry now holds each repo with its held data. small is computed with the rest of the row when it is read, and the hot pulls are still resolved per reader.
• Tests: tsc --noEmit reports no errors. I ran the 79 test files with held or flow in their names, and all 488 tests pass. I did not run the full suite, since scripts/fast.ts said the diff is too broad to name and the forge runs the full check on the push.

A held card for a hot-file hold now lists the open pulls on that file after its sentence. The earlier check had failed because this branch didn't merge cleanly into main.

The conflict. docs/design/areas/boards.md clashed with main's new "Verified after landing (#7786)" section. I merged main into the branch and kept both sections: the .held-pulls one from this pull and the verified one from main. Nothing else conflicted.

What I ran. The gate script passes, tsc --noEmit is clean, and the 7 tests across the 2 test files this pull changes pass. I did not run the full suite. scripts/fast.ts said the merge touches too many files to name tests, so I ran the changed test files myself and left the full check to the forge.

I didn't change any code in this fix, and I didn't look at the preview.

To see it, open a held card for a hot-file hold on the pull's preview; the open pulls on that file are listed after the hold's sentence.