fix(launcher): self-heal never fired under pnpm 11 by colinhacks · Pull Request #649 · nubjs/nub

colinhacks · 2026-08-01T02:22:27Z

leadsToUs() scanned a cmd-shim's quoted paths with /"([^"]+)"/g. That class cannot match an empty pair, so pnpm 11's exe="" / msys="" header lines each consumed one quote and re-paired every following quote off-by-one. The target token was never produced, so healPathEntry skipped the entry.

Every nub call under pnpm 11 kept paying the ~50ms node hop instead of the ~1-4ms sh trampoline. pnpm 10 has no empty assignment in its template and was unaffected — which is why the matrix stayed green.

Silent perf regression, not a crash: nothing failed, the fast path just never engaged.

Changes

widen the class to [^"]*
prefer pnpm >=11's own # cmd-shim-target=<abs> trailer when present — the shim declaring its own target, so it can't drift with template churn the way quote-scraping does
add a pnpm11 fixture style carrying both hazards, copied from a real pnpm@11.18.0 shim, and sweep it in the matrix

Verification

Fixture matrix: 26 ok / 5 failed before, 31 ok / 0 failed after. The new style fails without the launcher change.

Real pnpm@11.18.0 add @nubjs/nub@0.6.0, two fresh installs differing only in launch.js:

stock  before=57240cbc after=57240cbc  -> not healed
fixed  before=50a7bfa8 after=3c18b071  -> HEALED

leadsToUs() scanned a cmd-shim's quoted paths with /"([^"]+)"/g. That class cannot match an empty pair, so pnpm 11's `exe=""` / `msys=""` header lines each consumed a single quote and re-paired every following quote off-by-one. The real target token was never produced, leadsToUs returned false, and healPathEntry skipped the entry — so every `nub` call under pnpm 11 kept paying the ~50ms node hop instead of the ~1-4ms sh trampoline. pnpm 10, whose template has no empty assignment, was unaffected, which is why the existing matrix stayed green. Silent perf regression, not a crash: nothing failed, the fast path just never engaged. - widen the class to [^"]* so empty pairs pair correctly - prefer pnpm >=11's own `# cmd-shim-target=<abs>` trailer when present; it is the shim declaring its own target, so it cannot drift with template churn the way quote-scraping does - add a `pnpm11` fixture style carrying both hazards (empty assignments + the trailer), copied from a real pnpm@11.18.0 shim, and sweep it in the matrix Verified two ways. Fixture matrix: 26 ok/5 failed before, 31 ok/0 failed after. Real `pnpm@11.18.0 add @nubjs/nub@0.6.0`, two fresh installs differing only in launch.js: stock leaves the shim byte-identical, fixed rewrites it to the sh trampoline.

vercel · 2026-08-01T02:22:34Z

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
nub	Ready	Preview	Aug 3, 2026 1:52am

Copilot

Pull request overview

Fixes a silent launcher self-heal regression under pnpm >=11 by making leadsToUs() resilient to pnpm 11’s empty quoted assignments and by preferring pnpm’s explicit cmd-shim-target declaration when present. This restores the fast healed trampoline path so repeated nub invocations don’t keep paying the Node hop cost.

Changes:

Update leadsToUs() quote scanning to tolerate empty "" pairs and prefer pnpm >=11’s # cmd-shim-target=... trailer.
Add a pnpm 11-style fixture shim (empty exe=""/msys="" plus cmd-shim-target) to the launcher test harness.
Expand the launcher matrix to sweep the new pnpm11 style for both the main and concurrency scenarios.

Reviewed changes

Copilot reviewed 2 out of 3 changed files in this pull request and generated no comments.

File	Description
tests/launcher/run-launcher-matrix.sh	Adds `pnpm11` to the per-Node sweep for block + concurrency scenarios.
tests/launcher/make-fixture.sh	Adds a pnpm >=11 cmd-shim fixture variant that reproduces the parse hazards and includes `cmd-shim-target`.
npm/nub/bin/launch.js	Fixes `leadsToUs()` parsing for pnpm 11 and prefers pnpm’s declared target when available.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

pullfrog

Important

The diagnosis and the regex fix are correct — I confirmed both against the real upstream template. But the new pnpm11 fixture passes with the buggy regex still in place, so the silent regression this PR fixes has no regression guard, and the added # cmd-shim-target= trust path widens the clobber guard past the invariant documented directly above it.

Reviewed changes — the leadsToUs() parse fix in the self-heal launcher, plus the fixture style and matrix sweep added to cover it.

Quote-scan widened to [^"]* — npm/nub/bin/launch.js:146. Correct, and it generalizes: I probed old vs new against the real @zkochan/cmd-shim template (5-branch exec chain, *WSL2* arm, cygpath call), not just the fixture — [^"]+ yields 0 clean $basedir target tokens, [^"]* yields 3.
New # cmd-shim-target= trust path — npm/nub/bin/launch.js:139-145. The premise checks out: that trailer is emitted by @zkochan/cmd-shim 9.0.6 (shimTarget() in src/index.ts), which is what pnpm v11.18.0's bin-linker resolves, and upstream exports isShimPointingAt to read it. Concern is with how much it is trusted, inline.
pnpm11 fixture style — tests/launcher/make-fixture.sh:80-106, swept in tests/launcher/run-launcher-matrix.sh:149,152. Carries both hazards, but see inline — carrying both is what defeats the coverage.

ℹ️ The harness README still documents two shim styles

tests/launcher/README.md is the documented entry point for this loop and was not updated for the third style. Line 17 gives the usage as make-fixture.sh [dest] [symlink|pnpm], and line 46 says the host scenarios cover "both shim shapes". Not anchorable inline — the file is not in the diff.

Technical details

# Update the launcher harness README for the `pnpm11` style

## Affected sites
- `tests/launcher/README.md:17` — usage string reads `[symlink|pnpm]`; omits `pnpm11`, and the sentence describing the styles ("`symlink` reproduces npm/bun/yarn's on-PATH shape; `pnpm` reproduces the cmd-shim shape") does not mention the pnpm 11 shape or why it is separate.
- `tests/launcher/README.md:46` — "both shim shapes, swept across Node versions" is now three shapes.

## Required outcome
- The README describes all three on-PATH entry shapes and names what the `pnpm11` style specifically guards (the empty-quote-pair parse hazard, and the `# cmd-shim-target=` trailer), so a future reader knows why it exists as a separate style rather than deleting it as redundant with `pnpm`.

ℹ️ Nitpicks

tests/launcher/make-fixture.sh:82-83 describes the shim as "copied from a real pnpm@11.18.0 add shim", but it collapses upstream's 5-branch exec chain to a 2-branch if/else and drops both the *WSL2* case arm and the cygpath call inside the *CYGWIN*|*MINGW*|*MSYS* arm. I verified the fix parses the full real template correctly, so there is no correctness gap — but the wording overstates fidelity, and a verbatim paste would make the fixture a stronger guard against the template churn the PR is worried about.
npm/nub/bin/launch.js:144 passes declared[1].trim() straight to realpathSync, whereas the quoted-token branch resolves relatives against the shim's own directory (path.join(basedir, p), line 150). A relative declaration would resolve against the invoking process's CWD instead. Harmless today since upstream writes an absolute src, but the two branches reading the same shim disagree on what a relative path means.

^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using Claude Opus ｜ 𝕏}

The trailer value is not guaranteed absolute. pnpm's own reader does path.resolve(path.dirname(shShim), target) (selfUpdate.ts) and carries a test named 'self-update resolves relative legacy shim targets from pnpmHomeDir' that writes a relative one. leadsToUs passed the raw value to realpathSync, which resolves a relative path against CWD. Severity is latent, not live: in a full pnpm 11 shim the quoted-path scan resolves the same target anyway, and the pnpm11rel fixture passes with or without this change. It bites only on pnpm's trailer-ONLY legacy shim shape, where no quoted path exists to fall back to — measured there as absolute-only -> ENOENT -> heal skipped, resolved -> match. That shape has no exec line so the harness cannot drive it; recorded as a comment instead. Adds a pnpm11rel fixture style pinning that a relative trailer does not break the normal path.

pullfrog

Important

The path.resolve change is correct and cleanly fixes the CWD-relative nitpick from the last round. But the pnpm11rel style it ships with cannot fail for it, or for the regex fix — I measured all three ablations green — so the delta adds two matrix legs per Node to CI for zero coverage. Both inline findings from the previous review are still open.

Reviewed changes — the delta since eeac070 resolves the # cmd-shim-target= trailer against the shim's directory and adds a fixture style pinning a relative trailer.

Trailer resolved against the shim dir — npm/nub/bin/launch.js:159. path.resolve(basedir, …) is the right call and makes the two branches of leadsToUs agree on what a relative path means, which is what the prior nitpick asked for. No-op on an absolute value, so real pnpm 11 shims are unaffected.
New pnpm11rel fixture style — tests/launcher/make-fixture.sh:82,90-94, swept in tests/launcher/run-launcher-matrix.sh:149,152. Same shim as pnpm11 with a relative trailer. See inline.
Justification comment expanded — npm/nub/bin/launch.js:143-156. Cites pnpm's own reader and concedes the new fixture does not exercise the resolve.

ℹ️ Nitpicks

npm/nub/bin/launch.js:143-145 — "The value is NOT guaranteed absolute" overstates what upstream supports. shimTarget(src) in @zkochan/cmd-shim is a pass-through of whatever src the caller hands cmdShim(), and pnpm's bin-linker always passes an absolute one (@pnpm/bins.linker: depDir = path.resolve(opts.modulesDir, alias); @pnpm/bins.resolver: binPath = path.join(pkgPath, binRelativePath)). The cited path.resolve(path.dirname(shShim), target) is hasLegacyHomeDirShim hardening pnpm's own migration detector, and the "dedicated test for the relative form" hand-writes its shim body with fs.writeFileSync — it is not a sample of generator output. The code is still worth keeping for branch consistency; the comment should say "defensive: no writer path emits a relative value" rather than implying the shape occurs.
npm/nub/bin/launch.js:139-156 — 18 comment lines for a 3-line branch, and 150-156 narrate a fixture ablation rather than design, invariant, or provenance. AGENTS.md's bar is sparse and dense; the first paragraph plus one sentence on why resolve and not bare realpathSync carries everything a future reader needs.
tests/launcher/README.md is now two styles behind, not one: line 17 still gives the usage as [symlink|pnpm] and line 46 still says the host scenarios cover "both shim shapes".

^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using Claude Opus ｜ 𝕏}

Two defects in the same area, both measured. leadsToUs read any regular-file PATH entry as utf8 and regexed it. A PATH `nub` is usually a PM shim, but it can be a REAL nub binary — the curl install puts one at ~/.nub/bin alongside an npm install. Measured on the shipped 45MB darwin-x64 binary: 379ms to read, 41,781 quoted matches, 17,355 realpath syscalls, 1135ms total, versus 0ms on an actual shim. Under a PM whose heal never lands that is paid on every call, so it compounds with the pnpm 11 bug this branch fixes. Every shim shape we handle is ~0.5-2KB and starts with `#!`, so a size cap plus a shebang probe rejects the binary in two syscalls. Guard verified to bail at 0ms. Separately, npm/nub/bin/nubx was committed 100644 while bin/nub is 100755, and that asymmetry propagates into the published tarball (confirmed in the 0.0.31 archive: bin/nub -rwxr-xr-x, bin/nubx -rw-r--r--). npm re-adds +x for bin-field entries on install so the practical blast radius is small, but the mode should not differ between two files of the same kind. Launcher matrix: 41 ok, 0 failed.

pullfrog

Important

Both new guards are correct, and I verified they reject no real shim shape — every POSIX on-PATH bin entry npm, pnpm 10/11, yarn classic, yarn berry, bun, and Corepack produce is either a symlink or a sub-1 KB #!/bin/sh script. But this commit's second advertised fix, the +x on bin/nubx, is not in the commit: nubx is still mode 100644 at 05f6700, blob-and-mode identical to main.

Reviewed changes — the delta since 067bbbe adds two early-return preconditions to leadsToUs() ahead of the shim parse.

Size cap before the read — npm/nub/bin/launch.js:144. st.size > 64 * 1024 rejects a real nub binary on $PATH using the lstat the caller already took, so the 45 MB slurp costs nothing extra. The bound matches MAX_BIN_SHIM_BYTES in vendor/aube/crates/aube-linker/src/sys.rs:973, where aube caps its own shim reader at the same constant for the same reason.
Shebang probe after the read — npm/nub/bin/launch.js:146. Verified safe by building fixtures and running the real tools: npm 10, pnpm 10 and 11, yarn classic 1.22, yarn berry 4.5 under both node-modules and pnp, bun 1.3 under both linkers, and Corepack enable. Every regular-file entry is #!/bin/sh at 880–944 B; everything else is a symlink, handled by the earlier branch. Berry's PnP mode writes no PATH entry at all. Launcher matrix stays 41 ok / 0 failed on node v24.18.0.
Incidental narrowing of the clobber guard — the probe also rejects the exact mode-0644, no-shebang, comment-only file from the open launch.js:139-145 thread; I re-measured it and it now returns false. It does not close that thread: a shebang-led file whose only mention of us is # cmd-shim-target=<our launcher> still returns true, and so does the \s-across-a-newline variant.

⚠️ The commit's `+x on bin/nubx` fix is not in the tree

05f6700 documents two defects and fixes one. git ls-tree HEAD npm/nub/bin/ returns nubx at 100644 with the same blob as origin/main, and git show --stat 05f6700 is one file, nine insertions, launch.js only — the chmod was never staged. Runtime impact is small exactly as the message says (nubx is a bin-field entry in npm/nub/package.json:26, so npm re-adds +x on install), but the mode asymmetry it describes still ships in the tarball, and the commit message permanently asserts a change a future reader will not find.

Technical details

# `npm/nub/bin/nubx` is still mode 0644; the commit message claims otherwise

## Affected sites
- `npm/nub/bin/nubx` — mode `100644` at head `05f6700`, identical blob and mode to `origin/main`. `npm/nub/bin/nub` is `100755`.
- `05f6700`'s commit message — paragraph 3 states the mode was corrected ("Separately, npm/nub/bin/nubx was committed 100644 while bin/nub is 100755"), but `git show --stat` for that commit lists only `npm/nub/bin/launch.js`.

## Required outcome
- Either the mode change lands (`chmod +x npm/nub/bin/nubx`, staged so git records `100755`), or the paragraph claiming it comes out of the commit message. The tree and its history must agree.
- Since the message is only editable while the PR is open, resolving this after merge means a force-push or a permanently inaccurate record.

## Open questions for the human
- Was the `chmod` intended for this PR at all, or did it get folded into the message from a separate local change? If it belongs elsewhere, dropping the paragraph is the cheaper fix.

ℹ️ Nitpicks

npm/nub/bin/launch.js:142-143 — "a size cap plus a shebang probe rejects the binary for two syscalls" conflates the two guards. The size cap reuses the lstat healPathEntry already took at line 208, so it rejects the 45 MB binary for zero extra syscalls; the shebang probe is what needs an open/read/close, and only for files that already passed the cap. "the size cap alone rejects it off the lstat we already have" states the actual mechanism.
npm/nub/bin/launch.js:137-165 — now ~29 comment lines carrying two one-line guards and a three-line branch. The measurement paragraph (137-143) earns its place as provenance; lines 159-165, which narrate which fixture does and does not exercise the resolve, are the ablation-log shape AGENTS.md asks to keep out of source comments.

^{｜ Fix it ➔ ｜ View workflow run ｜ Using Claude Opus ｜ 𝕏}

The pnpm 11 fix this branch lands had no regression guard. leadsToUs tries the `# cmd-shim-target=` trailer before the quote scan and returns on a match, so a fixture carrying both hazards -- which is exactly what a real pnpm 11 shim is -- still heals when either mechanism alone regresses. Measured by ablation: reverting `[^"]*` to `[^"]+` left the whole matrix green, on the synthetic fixture and on a shim written by real pnpm@11.18.0 alike. That is how the bug shipped green once. Replace pnpm11/pnpm11rel with four styles. pnpm11 is now byte-identical to real @zkochan/cmd-shim 9.0.6 output (5-branch exec chain, *WSL2* arm, cygpath call), verified by diffing against the generator. `scan` withholds the trailer so only the quote scan can match; `decl`/`declrel` assemble the target into an UNQUOTED variable so no quoted token resolves to us and only the trailer branch can match, the second with a relative value. The comment claiming that shape could not be driven end-to-end was wrong -- it runs fine, it just needs an unquoted target. Two guards on trusting the declaration, both because healPathEntry's rename has no backup. Require an `exec` in the body: a file whose only mention of our launcher is `# cmd-shim-target=<us>` was being renamed over, at mode 0644 and 0755 alike, which contradicted the invariant stated directly above the branch. And `[ \t]*` instead of `\s*`, which spans newlines and paired a bare `#` line with a following key line. Both now have a foreign-nub negative test. Verified against real pnpm 11.18.0 and 10.20.0 (regular-file shims, both heal) and npm 11, yarn 1.22, bun (symlinks, which never reach this branch), so the corroboration rejects no real shim shape. Concurrency now sweeps only symlink/pnpm/pnpm11 -- the polyglot race it guards is in the heal write, identical whichever route matched, so a leg per parse style cost 200 forks for nothing. Also chmod +x npm/nub/bin/nubx: 05f6700's message documented that change but the mode was never staged, so the tree still had 100644 against bin/nub's 100755. Ablation, each launcher change reverted independently: [^"]* -> [^"]+ -> scan red drop the trailer branch -> decl + declrel red path.resolve -> bare value -> declrel red drop the exec corroboration -> foreign "declares our target, never execs" red [ \t]* -> \s* -> foreign "bare # then newline then key" red Matrix: 285 ok, 0 failed across node 18.19.0, 20.19.0, 22.14.0, 22.15.0, 26.5.0.

pullfrog

ℹ️ No critical issues — two minor coverage suggestions inline.

The fixture restructure genuinely closes all three prior threads, and I verified it the way the new README asks to be verified: by ablation, not by a green run. All five reverts the README claims take a leg red actually do (measured on node v24.18.0). Both remaining suggestions are about coverage of the harness itself, not about the fix.

Reviewed changes — the delta since 05f6700 replaces the un-failable pnpm11/pnpm11rel pair with mechanism-isolating styles, adds corroboration to the trailer branch, and lands the +x the previous commit's message promised.

Trailer read gated on exec and tightened to [ \t]* — npm/nub/bin/launch.js:164-165. Both guards the prior thread asked for. I checked the gate against the generator rather than the fixture: npm/nub/bin/nub is #!/usr/bin/env node, so @zkochan/cmd-shim resolves prog = 'node', and @pnpm/bins.linker passes nodeExecPath only when engines.runtime is set (nub declares plain engines.node) — so the shim always takes generateShShim's shLongProg branch, which has carried exec since 2020. A real trailer-without-exec window did exist (@zkochan/cmd-shim 9.0.0–9.0.2, fixed by e4f6173a) but only on the branch a shebang'd target cannot reach.
Doc comment now states the actual precondition — npm/nub/bin/launch.js:121-126. The stale "rejects comment-only mentions" invariant is gone, replaced by what the code enforces. "as pnpm's own reader does" also checks out: pnpm's readShimTarget uses \s*, and the clause attaches to the path.resolve step, which does match.
pnpm11 is now verbatim upstream — tests/launcher/make-fixture.sh:107-145. Verified empirically, not by reading the generator: packed @zkochan/cmd-shim@9.0.6, ran cmdShim() against a node-shebang target, and string-compared — 1067 bytes each, identical, zero per-line mismatches. uname -a, the basedir_win/exe/msys ordering, both case arms' indentation, the double space in exec node "<target>", and the trailer as the last line all match. The missing exit $? is correct: upstream emits it only on the shLongProg-unset branch. The fidelity nitpick from eeac070 is closed.
Three derived styles that can each go red alone — tests/launcher/make-fixture.sh:102-166, swept at tests/launcher/run-launcher-matrix.sh:173. Every claim in the README's revert table holds: [^"]+ → scan red (4 failures, scan only); trailer branch dropped → decl + declrel red (8); path.resolve dropped → declrel red (4); exec gate dropped → the "never execs" probe red; \s* → the "bare # then newline" probe red. Full matrix at head is 57 ok / 0 failed.
run_foreign extended to three probes — tests/launcher/run-launcher-matrix.sh:143-163. Both new probes are real negative tests, not decoration.
npm/nub/bin/nubx is 0755 — the diff records old mode 100644 / new mode 100755, closing the 05f6700 body finding.

ℹ️ Nitpicks

tests/launcher/run-launcher-matrix.sh:2-3 — the script's own header still says it exercises "both package-manager bin-shim shapes (symlink: npm/bun/yarn; cmd-shim: pnpm)". That is six styles now, and the scenario list on lines 9-15 doesn't mention that three of them exist to isolate a parse mechanism. The README got this treatment; the script header is the same staleness one file over. Not anchorable — the lines aren't in a diff hunk.

^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using Claude Opus ｜ 𝕏}

…style Two gaps from review, both instances of the thing this branch is about — a check that cannot go red, and a green leg that pins nothing. The `#!` precondition in leadsToUs had no probe. All three foreign bodies opened with `#!/bin/sh`, so `if (!body.startsWith("#!")) return false;` could be deleted with the matrix still at 57 ok. That check is a clobber guard, not only the perf guard the size cap is: it is what rejects a no-shebang file carrying `# cmd-shim-target=<ours>`. Add a fourth foreign probe for exactly that shape at mode 0644, and give it a row in the README revert table. The size cap stays unpinned and the README now says so — a 64KB fixture costs more than it is worth. make-fixture.sh fell through to the pnpm 10 shim for any unrecognized style, so `make-fixture.sh /tmp/x typo-style` exited 0 and wrote the wrong body. The matrix spells all six style names by hand, so a misspelling there would have bought a green leg guarding nothing. Name the arm `pnpm)` and exit 2 on anything unknown. Also refresh the matrix script's own header, which still described "both package-manager bin-shim shapes" and did not mention that three of the six styles exist to isolate one parse mechanism. Ablation, each guard reverted independently: drop the `#!` precondition -> foreign "no shebang, declares our target" red [^"]* -> [^"]+ -> scan red drop the trailer branch -> decl + declrel red path.resolve -> bare value -> declrel red drop the exec corroboration -> foreign "declares our target, never execs" red [ \t]* -> \s* -> foreign "bare # then newline then key" red Unknown style now exits 2. Matrix: 290 ok, 0 failed across node 18.19.0, 20.19.0, 22.14.0, 22.15.0, 26.5.0.

pullfrog

✅ No new issues found.

The delta is harness-only and it closes both remaining threads. I verified it the way the README asks — by ablation, not by a green run: the new no-shebang probe genuinely pins if (!body.startsWith("#!")) return false; (58 ok / 0 failed at head; delete that line and it is 57 ok / 1 failed, the sole failure being foreign CLOBBERED: no shebang, declares our target). Every launcher change in this PR now has a leg that can go red on its own.

Reviewed changes — the delta since 04f25b2 pins the shebang precondition with a negative probe, makes an unknown fixture style a hard error, and finishes the doc staleness sweep.

Fourth run_foreign probe — tests/launcher/run-launcher-matrix.sh:172-176. A mode-0644 body that declares our target and contains exec but has no shebang. Measured red without the guard and green with it, on node v24.18.0.
make-fixture.sh no longer aliases a typo to the pnpm 10 shape — tests/launcher/make-fixture.sh:167,176-184. The old catch-all is now an explicit pnpm) arm and *) errors: bash make-fixture.sh /tmp/x typo-style → exit 2 plus the six valid names on stderr, while declrel still exits 0. Every caller (run-launcher-matrix.sh:56,112,141, docker-non-owner.sh:17) passes a valid style explicitly and the $2 default is symlink, so nothing breaks. run-launcher-matrix.sh is set -uo pipefail rather than -e, so a bad style does not abort the sweep — but the stderr line prints and the dependent legs go red, which is loud enough.
Sixth row in the README revert table — tests/launcher/README.md:47-49, with the reason the #! check belongs on a list of clobber guards while the size cap does not. All six rows now hold; I have measured every one across this PR's reviews.
Matrix-script header rewritten — tests/launcher/run-launcher-matrix.sh:2-12,21-22. The "both bin-shim shapes" claim is gone, the derived styles are explained where a reader meets them, and the foreign scenario line names the "declares our target without dispatching to it" cases. This closes the nitpick from the last round.

^{｜ View workflow run ｜ Using Claude Opus ｜ 𝕏}

colinhacks · 2026-08-05T08:08:12Z

Shipped in v0.7.0: https://github.com/nubjs/nub/releases/tag/v0.7.0

Copilot AI review requested due to automatic review settings August 1, 2026 02:22

Copilot started reviewing on behalf of colinhacks August 1, 2026 02:22 View session

vercel Bot deployed to Preview August 1, 2026 02:24 View deployment

Copilot AI reviewed Aug 1, 2026

View reviewed changes

pullfrog Bot reviewed Aug 1, 2026

View reviewed changes

Comment thread tests/launcher/make-fixture.sh Outdated

Comment thread npm/nub/bin/launch.js Outdated

vercel Bot deployed to Preview August 1, 2026 02:38 View deployment

pullfrog Bot reviewed Aug 1, 2026

View reviewed changes

Comment thread tests/launcher/make-fixture.sh Outdated

vercel Bot deployed to Preview August 1, 2026 02:58 View deployment

pullfrog Bot reviewed Aug 1, 2026

View reviewed changes

vercel Bot deployed to Preview August 3, 2026 00:41 View deployment

pullfrog Bot reviewed Aug 3, 2026

View reviewed changes

Comment thread tests/launcher/run-launcher-matrix.sh

Comment thread tests/launcher/make-fixture.sh Outdated

vercel Bot deployed to Preview August 3, 2026 01:52 View deployment

pullfrog Bot reviewed Aug 3, 2026

View reviewed changes

colinhacks merged commit 2a1f1e7 into main Aug 3, 2026
47 checks passed

colinhacks deleted the fix-pnpm11-heal branch August 3, 2026 02:12

Navigation Menu

Uh oh!

fix(launcher): self-heal never fired under pnpm 11 - #649

fix(launcher): self-heal never fired under pnpm 11#649
colinhacks merged 5 commits into
mainfrom
fix-pnpm11-heal

colinhacks commented Aug 1, 2026

Uh oh!

vercel Bot commented Aug 1, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

pullfrog Bot left a comment

Uh oh!

Uh oh!

Uh oh!

pullfrog Bot left a comment

Uh oh!

Uh oh!

pullfrog Bot left a comment

Uh oh!

pullfrog Bot left a comment

Uh oh!

Uh oh!

Uh oh!

pullfrog Bot left a comment

Uh oh!

Uh oh!

colinhacks commented Aug 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

colinhacks commented Aug 1, 2026

Changes

Verification

Uh oh!

vercel Bot commented Aug 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

pullfrog Bot left a comment

Choose a reason for hiding this comment

ℹ️ The harness README still documents two shim styles

ℹ️ Nitpicks

Uh oh!

Uh oh!

Uh oh!

pullfrog Bot left a comment

Choose a reason for hiding this comment

ℹ️ Nitpicks

Uh oh!

Uh oh!

pullfrog Bot left a comment

Choose a reason for hiding this comment

⚠️ The commit's +x on bin/nubx fix is not in the tree

ℹ️ Nitpicks

Uh oh!

pullfrog Bot left a comment

Choose a reason for hiding this comment

ℹ️ Nitpicks

Uh oh!

Uh oh!

Uh oh!

pullfrog Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

colinhacks commented Aug 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

vercel Bot commented Aug 1, 2026 •

edited

Loading

⚠️ The commit's `+x on bin/nubx` fix is not in the tree