Skip to content

NapkinStack v0.6.1 → v0.6.2 - #31

Merged
napkinstack-agent[bot] merged 1 commit into
mainfrom
nstack/update-v0.6.2
Sep 24, 2026
Merged

napkinstack-agent[bot] merged 1 commit into
mainfrom
nstack/update-v0.6.2

Conversation

@napkinstack-agent

Copy link
Copy Markdown
Contributor

NapkinStack v0.6.1 → v0.6.2

Taken at the same boundary as #29, and deliberately before cycle 03 is accepted (#30): once that
merges, the project is inside a cycle and the next check is owed at 2026-10-15.


Module

None. Two lines: .copier-answers.yml and the install command in README.md.

  • This pull request touches a single module — n/a, it touches none
  • ⚠️ Cross-module

Deliverable

Out of cycle: a framework update.

What it changes

-_commit: v0.6.1
+_commit: v0.6.2
-uv tool install "napkinstack==0.6.1" --with-executables-from pre-commit
+uv tool install "napkinstack==0.6.2" --with-executables-from pre-commit

That is the whole diff, and it is what the release said it would be. v0.6.2's changelog states
"In your project — nothing":
skeleton/ is unchanged since v0.6.1, so nstack update lays down a
branch that bumps the recorded version and merges no file. No conflict was reported and git grep
finds no marker.

Checked rather than taken on trust. The nine files that changed between the two tags are all in
the engine or in its own governance documents; none is under skeleton/, and none is
plan.py or pull_request.py — so nothing already merged becomes non-compliant under the new
judge.

What the engine gains, and why it is worth a boundary

  • A changelog is now the source of a release. The publishing workflow refuses a version whose
    section is missing and posts that section as the tag's release page.
  • nstack update says where to read it, and names every version crossed, not the target
    alone. This landed for us immediately: the command printed the changelog URL and "every section
    above v0.6.1".
  • A refusal carrying an errno names the filesystem instead of reporting every OSError as
    "Template unreachable" with an action about --source, --ref and the network.

The second one is the reason not to wait. Checking this boundary turned up something worth knowing:
v0.6.1's changelog says it changed .gitignore and that a project with its own agent ignore rules
would conflict — and our v0.6.1 update (#29) produced two lines and no conflict. That could have
been a skeleton change silently dropped. It was not: our .gitignore is the v0.6.1 skeleton's plus
our own two .coverage lines, the D58 rules are all present, and .claude/settings.json and
.claude/hooks/ are deliberately not ignored. The conflict never happened because #4 had already
written the same rules. From now on the tool reports that instead of leaving it to be
reconstructed by hand.

The sequence, kept and now twice recorded

Cycle 02 closed (#27) → framework to v0.6.1 (#29) → this → cycle 03 accepted (#30). A cycle
accepted under a framework version the project has not recorded would be a cycle judged by
something the repository does not declare.

nstack update was run from main this time, so unlike #29 there was no feature-branch commit to
replant.


Oracle

  • The success criterion was written before the implementation — nstack doctor's L1, which
    compares the installed CLI against the version recorded in .copier-answers.yml
  • It was seen failing, for the right reason — L1 went red between the install and the
    update during the v0.6.1 round, exactly as the handover warns
  • It passes now — nstack doctor: 0 gaps, L1 to L7 OK, judged by 0.6.2 with the project
    recorded at v0.6.2

Kind of oracle: fitness function · nstack doctor

Review budget

  • Lines changed: 2
  • Files touched: 2
  • Within budget

Contracts

  • No contract affected

Definition of Done

  • Oracle green — nstack doctor 0 gaps
  • Lint, format, types — nstack check screening exit 0 on this branch
  • Unit tests — 81 passed on this branch
  • Fitness functions — nstack fitness exit 0, compliant
  • Everything else — n/a, no code
  • Affected documentation updated — the README's install line is in the diff
  • Diff self-reviewed line by line

Test sheet

Not required: this pull request changes no module.


Summary

DONE
The project is recorded at NapkinStack v0.6.2 and the README installs it.

VERIFIED
nstack doctor 0 gaps with L1 to L7 OK; nstack fitness exit 0; nstack check screening exit 0;
81 tests pass on this branch. The claim that nothing changes in the project was checked against the
tag diff, not only against the changelog.

ASSUMED
Nothing. The changelog's "In your project — nothing" was confirmed by listing the files that
changed between the tags.

NOT VERIFIED
The thirteen GitHub settings: nstack doctor reports them NOT VERIFIED because the shell had no
token — what cannot be read is never reported as guarded. doctor exits non-zero in that state,
so the gap count is what to read. Nothing here touches those settings.

RISKS
None identified. The rollback is one commit. The framework has published three versions in four
days, so the next check is owed at the 2026-10-15 boundary and not before — chasing releases
mid-cycle is what the rule exists to prevent.


Signals to report

  • I had to look at another module's implementation
  • A kernel or playbook rule got in my way for no good reason
  • A quality gate blocked without improving quality
  • A rule applied by hand would be worth automating

The signal raised in #29 — that nstack update branches from the current HEAD without saying so —
still stands, and this round avoided it by checking out main first rather than by the tool
changing.

@napkinstack-agent
napkinstack-agent Bot merged commit 5afa957 into main Sep 24, 2026
7 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants