Tags: pullfrog/pullfrog

release: bump action to 0.1.60

release: publish to npm before moving the v0 tag, and bump to 0.1.59

the 0.1.58 release moved v0 to an unpublished version and took every
consumer down. the job pushed tags first, then created the GitHub
release, then published to npm — so when actions/create-release@v1 hit a
transient GitHub 5xx, v0 already pointed at 0.1.58 while npm still
served 0.1.57. the action bakes pullfrog@^<own version> into its npx
spec, so every pullfrog/pullfrog@v0 run resolved ETARGET.

- publish to npm first, then confirm the registry is serving the version,
  and only then move the tags. v0 can no longer advance past npm.
- gate on whether npm serves the version rather than on tag existence.
  the old guard was self-defeating: the job pushes the tag itself, so a
  re-run saw its own tag and skipped the publish it still owed, which
  made a half-finished release unrecoverable.
- replace actions/create-release@v1 (archived by GitHub in 2021, and the
  step that actually failed) with gh release create, and run it last so
  its failure costs a re-run rather than a consumer outage.
- make every step idempotent so the job converges on re-run.

release: bump action to 0.1.58

release: bump action to 0.1.60

release: bump action to 0.1.57

release: bump action to 0.1.56

hotfix: revert the model-introspection dirty window, release 0.1.55

0.1.54 failed 100% of production runs. Within ~15 minutes of publish, 24 of 25
runs across 11 distinct accounts died about one second in, every one with:

  git diff --name-only HEAD failed (exit 129):
  warning: Not a git repository. Use --no-index to compare two paths outside a
  working tree

#1213 added a `dirtyTrackedPaths()` snapshot around the `opencode models`
introspection to fix #1151. `dirtyTrackedPaths()` THROWS on a non-zero git exit,
and on the runner that git call exits 129 at this point even though the checkout
step demonstrably succeeded moments earlier (confirmed in the logs of
nubjs/nub, cardvex/bedrock, membranehq/platform and pullfrog/app). So the run
died before the agent ever started.

Every workflow resolves `pullfrog@^0.1.x` through npx, so the caret range moved
the whole fleet onto the broken version the instant it hit npm — nobody changed
a setting.

Reverting to known-good 0.1.53 behavior rather than guarding the call, because
the cwd mechanism is not yet explained and a 100% outage outranks the silent
diff-loss of #1151. #1151 still needs a fix; it must not be one that can abort
the run.

release: bump action to 0.1.54

release: bump action to 0.1.53

release: bump action to 0.1.52