fix(fork): a sync fold keeps the squash-landed fixup's PR link by donjor-agent[bot] · Pull Request #1620 · RSI-Software/t3code-hyprws

donjor-agent · 2026-10-04T03:13:23Z

🤖 claude/anthropic claude-opus-5-5 (high)

Problem

The sync rebase folds a squash-landed fixup! <owner subject> (#N) into its owner as git's fixup, which keeps only the owner's message.
So (#N) is lost, breaking the #1491 rule that every PR link survives every fold.

Change

Sync renders each folded owner's message first
Adds the Squashes: list fork:fold writes
Lists each fixup's PR link there
Keeps trailers verbatim, Co-authored-by: too
Sequence editor amends the owner with it
Link guard refuses a tip that lost a link

A fixup citing no PR still folds as plain git fixup, its owner's message untouched.
fork-sync.md now names this as the sync's only reword.

Guard scope: it also refuses a sync where git silently drops a linked fork commit.
That matches the sync's rule that it never drops a commit.

Scope and approval

Closes #1592.

Sync: a folded fixup! <subject> (#N) keeps RSI-Software/t3code-hyprws#N on the owner's message
Guard: the sync refuses a rebased tip that drops a PR link

Verification

test	result
Fold of `fixup! … (#1600)`	link kept as member line, trailers verbatim
Tip with the link dropped	refused
Real rebased tip	passes the guard

command	exit
`vp test run`, 3 touched test files, 127 tests	0
scripts `tsc --noEmit`	0
`vp run fork:ci`, scripts 739 tests	0

Not run: a real sync with a linked fixup on trunk; the next one proves it.

Claude Opus 5.5 in Claude Code, inside T3 Code; implemented by a Pi child agent.

Fork trailers

Fork-Domain: fork-meta
Fork-Tier: core
Fork-Upstreamable: no

The sync rebase folded each fixup! <subject> (#N) with git's fixup action, dropping the link the marker carried. The sequence editor now appends one exec per owner that amends the folded message with the Squashes: rendering fork:fold writes, and the rebase refuses a rebased tip whose link guard finds a dropped link (#1592). Fork-Domain: fork-meta Fork-Tier: core Fork-Upstreamable: no

> 🤖 claude/anthropic claude-opus-5-5 (high) ## Problem The sync rebase folds a squash-landed `fixup! <owner subject> (#N)` into its owner as git's `fixup`, which keeps only the owner's message. So `(#N)` is lost, breaking the #1491 rule that every PR link survives every fold. ## Change 1. Sync renders each folded owner's message first 2. Adds the `Squashes:` list `fork:fold` writes 3. Lists each fixup's PR link there 4. Keeps trailers verbatim, `Co-authored-by:` too 5. Sequence editor amends the owner with it 6. Link guard refuses a tip that lost a link A fixup citing no PR still folds as plain git `fixup`, its owner's message untouched. `fork-sync.md` now names this as the sync's only reword. **Guard scope:** it also refuses a sync where git silently drops a linked fork commit. That matches the sync's rule that it never drops a commit. ## Scope and approval Closes #1592. - [ ] **Sync:** a folded `fixup! <subject> (#N)` keeps RSI-Software/t3code-hyprws#N on the owner's message - [ ] **Guard:** the sync refuses a rebased tip that drops a PR link ## Verification | test | result | | --- | --- | | Fold of `fixup! … (#1600)` | link kept as member line, trailers verbatim | | Tip with the link dropped | refused | | Real rebased tip | passes the guard | | command | exit | | --- | --- | | `vp test run`, 3 touched test files, 127 tests | 0 | | scripts `tsc --noEmit` | 0 | | `vp run fork:ci`, scripts 739 tests | 0 | Not run: a real sync with a linked fixup on trunk; the next one proves it. Claude Opus 5.5 in Claude Code, inside T3 Code; implemented by a Pi child agent. ## Fork trailers Fork-Domain: fork-meta Fork-Tier: core Fork-Upstreamable: no

## Problem The sync to `v0.1.1-nightly.20261020.3386` fails after a clean rebase: ```text the rebased tip dropped 1 pull request link(s): #1801: cited by fe075c9 ..., by no new commit ``` #1801 backported upstream `2532e64099` exactly; both share patch-id `4221c8a56228`. The rebase drops it correctly, as its body said it would. The link guard from #1620 has no exemption for that drop, so no sync can pass. ## Change `refuseDroppedLinks` skips old commits that `git cherry <target> <old>` marks `-`, meaning the target already carries the patch. Every other dropped link still refuses. ## Scope and approval Maintainer-authorized: unblocks the requested fork sync. ## Verification | Check | Result | | --- | --- | | New test, without fix | fails | | New test, with fix | passes | | `scripts/fork-sync.test.ts` | 69 passed | | Guard on the real sync shas | passes | | `vp run fork:ci` | exit 0 | Written by Claude Opus 5.5 in Claude Code, hosted by T3 Code. ## Fork trailers Fork-Domain: fork-meta Fork-Tier: bugfix Fork-Upstreamable: no Co-authored-by: donjor <[email protected]>

> 🤖 claude/anthropic claude-opus-5-5 (high) ## Problem The sync rebase folds a squash-landed `fixup! <owner subject> (#N)` into its owner as git's `fixup`, which keeps only the owner's message. So `(#N)` is lost, breaking the #1491 rule that every PR link survives every fold. ## Change 1. Sync renders each folded owner's message first 2. Adds the `Squashes:` list `fork:fold` writes 3. Lists each fixup's PR link there 4. Keeps trailers verbatim, `Co-authored-by:` too 5. Sequence editor amends the owner with it 6. Link guard refuses a tip that lost a link A fixup citing no PR still folds as plain git `fixup`, its owner's message untouched. `fork-sync.md` now names this as the sync's only reword. **Guard scope:** it also refuses a sync where git silently drops a linked fork commit. That matches the sync's rule that it never drops a commit. ## Scope and approval Closes #1592. - [ ] **Sync:** a folded `fixup! <subject> (#N)` keeps RSI-Software/t3code-hyprws#N on the owner's message - [ ] **Guard:** the sync refuses a rebased tip that drops a PR link ## Verification | test | result | | --- | --- | | Fold of `fixup! … (#1600)` | link kept as member line, trailers verbatim | | Tip with the link dropped | refused | | Real rebased tip | passes the guard | | command | exit | | --- | --- | | `vp test run`, 3 touched test files, 127 tests | 0 | | scripts `tsc --noEmit` | 0 | | `vp run fork:ci`, scripts 739 tests | 0 | Not run: a real sync with a linked fixup on trunk; the next one proves it. Claude Opus 5.5 in Claude Code, inside T3 Code; implemented by a Pi child agent. ## Fork trailers Fork-Domain: fork-meta Fork-Tier: core Fork-Upstreamable: no

## Problem The sync to `v0.1.1-nightly.20261020.3386` fails after a clean rebase: ```text the rebased tip dropped 1 pull request link(s): #1801: cited by fe075c9 ..., by no new commit ``` #1801 backported upstream `2532e64099` exactly; both share patch-id `4221c8a56228`. The rebase drops it correctly, as its body said it would. The link guard from #1620 has no exemption for that drop, so no sync can pass. ## Change `refuseDroppedLinks` skips old commits that `git cherry <target> <old>` marks `-`, meaning the target already carries the patch. Every other dropped link still refuses. ## Scope and approval Maintainer-authorized: unblocks the requested fork sync. ## Verification | Check | Result | | --- | --- | | New test, without fix | fails | | New test, with fix | passes | | `scripts/fork-sync.test.ts` | 69 passed | | Guard on the real sync shas | passes | | `vp run fork:ci` | exit 0 | Written by Claude Opus 5.5 in Claude Code, hosted by T3 Code. ## Fork trailers Fork-Domain: fork-meta Fork-Tier: bugfix Fork-Upstreamable: no Co-authored-by: donjor <[email protected]>

> 🤖 claude/anthropic claude-opus-5-5 (high) ## Problem The sync rebase folds a squash-landed `fixup! <owner subject> (#N)` into its owner as git's `fixup`, which keeps only the owner's message. So `(#N)` is lost, breaking the #1491 rule that every PR link survives every fold. ## Change 1. Sync renders each folded owner's message first 2. Adds the `Squashes:` list `fork:fold` writes 3. Lists each fixup's PR link there 4. Keeps trailers verbatim, `Co-authored-by:` too 5. Sequence editor amends the owner with it 6. Link guard refuses a tip that lost a link A fixup citing no PR still folds as plain git `fixup`, its owner's message untouched. `fork-sync.md` now names this as the sync's only reword. **Guard scope:** it also refuses a sync where git silently drops a linked fork commit. That matches the sync's rule that it never drops a commit. ## Scope and approval Closes #1592. - [ ] **Sync:** a folded `fixup! <subject> (#N)` keeps RSI-Software/t3code-hyprws#N on the owner's message - [ ] **Guard:** the sync refuses a rebased tip that drops a PR link ## Verification | test | result | | --- | --- | | Fold of `fixup! … (#1600)` | link kept as member line, trailers verbatim | | Tip with the link dropped | refused | | Real rebased tip | passes the guard | | command | exit | | --- | --- | | `vp test run`, 3 touched test files, 127 tests | 0 | | scripts `tsc --noEmit` | 0 | | `vp run fork:ci`, scripts 739 tests | 0 | Not run: a real sync with a linked fixup on trunk; the next one proves it. Claude Opus 5.5 in Claude Code, inside T3 Code; implemented by a Pi child agent. ## Fork trailers Fork-Domain: fork-meta Fork-Tier: core Fork-Upstreamable: no

## Problem The sync to `v0.1.1-nightly.20261020.3386` fails after a clean rebase: ```text the rebased tip dropped 1 pull request link(s): #1801: cited by fe075c9 ..., by no new commit ``` #1801 backported upstream `2532e64099` exactly; both share patch-id `4221c8a56228`. The rebase drops it correctly, as its body said it would. The link guard from #1620 has no exemption for that drop, so no sync can pass. ## Change `refuseDroppedLinks` skips old commits that `git cherry <target> <old>` marks `-`, meaning the target already carries the patch. Every other dropped link still refuses. ## Scope and approval Maintainer-authorized: unblocks the requested fork sync. ## Verification | Check | Result | | --- | --- | | New test, without fix | fails | | New test, with fix | passes | | `scripts/fork-sync.test.ts` | 69 passed | | Guard on the real sync shas | passes | | `vp run fork:ci` | exit 0 | Written by Claude Opus 5.5 in Claude Code, hosted by T3 Code. ## Fork trailers Fork-Domain: fork-meta Fork-Tier: bugfix Fork-Upstreamable: no Co-authored-by: donjor <[email protected]>

> 🤖 claude/anthropic claude-opus-5-5 (high) ## Problem The sync rebase folds a squash-landed `fixup! <owner subject> (#N)` into its owner as git's `fixup`, which keeps only the owner's message. So `(#N)` is lost, breaking the #1491 rule that every PR link survives every fold. ## Change 1. Sync renders each folded owner's message first 2. Adds the `Squashes:` list `fork:fold` writes 3. Lists each fixup's PR link there 4. Keeps trailers verbatim, `Co-authored-by:` too 5. Sequence editor amends the owner with it 6. Link guard refuses a tip that lost a link A fixup citing no PR still folds as plain git `fixup`, its owner's message untouched. `fork-sync.md` now names this as the sync's only reword. **Guard scope:** it also refuses a sync where git silently drops a linked fork commit. That matches the sync's rule that it never drops a commit. ## Scope and approval Closes #1592. - [ ] **Sync:** a folded `fixup! <subject> (#N)` keeps RSI-Software/t3code-hyprws#N on the owner's message - [ ] **Guard:** the sync refuses a rebased tip that drops a PR link ## Verification | test | result | | --- | --- | | Fold of `fixup! … (#1600)` | link kept as member line, trailers verbatim | | Tip with the link dropped | refused | | Real rebased tip | passes the guard | | command | exit | | --- | --- | | `vp test run`, 3 touched test files, 127 tests | 0 | | scripts `tsc --noEmit` | 0 | | `vp run fork:ci`, scripts 739 tests | 0 | Not run: a real sync with a linked fixup on trunk; the next one proves it. Claude Opus 5.5 in Claude Code, inside T3 Code; implemented by a Pi child agent. ## Fork trailers Fork-Domain: fork-meta Fork-Tier: core Fork-Upstreamable: no

## Problem The sync to `v0.1.1-nightly.20261020.3386` fails after a clean rebase: ```text the rebased tip dropped 1 pull request link(s): #1801: cited by fe075c9 ..., by no new commit ``` #1801 backported upstream `2532e64099` exactly; both share patch-id `4221c8a56228`. The rebase drops it correctly, as its body said it would. The link guard from #1620 has no exemption for that drop, so no sync can pass. ## Change `refuseDroppedLinks` skips old commits that `git cherry <target> <old>` marks `-`, meaning the target already carries the patch. Every other dropped link still refuses. ## Scope and approval Maintainer-authorized: unblocks the requested fork sync. ## Verification | Check | Result | | --- | --- | | New test, without fix | fails | | New test, with fix | passes | | `scripts/fork-sync.test.ts` | 69 passed | | Guard on the real sync shas | passes | | `vp run fork:ci` | exit 0 | Written by Claude Opus 5.5 in Claude Code, hosted by T3 Code. ## Fork trailers Squashes: - 4f2343c fix(fork-sync): an exact upstream backport may drop its PR link (#1805) (#1805) - eb10567 fixup! fix(fork-sync): an exact upstream backport may drop its PR link (#1805) (#1805) Fork-Domain: fork-meta Fork-Tier: bugfix Fork-Upstreamable: no