[StevenTCramer]

Summary

TimeWarp.Nuru.DevCli 4.0.0-beta.11 supplies clean, check-version, release, and self-install. This repo keeps build, test, pack, format, verify-samples, and workflow. Release mode promotes Packages-* and pushes TimeWarp.Builder to nuget.org. It does not rebuild. <Version> is still 1.1.0. No tag and no GitHub Release were created.

dev self-install AOT-publishes because tools/dev-cli/Directory.Build.props sets GenerateDocumentationFile to false, so the publish no longer looks for dev.xml. Amuru stays at 2.0.0: 3.0.0-beta.1 is net11.0 only, and the git hooks restore that package. DevCli 4.0.0-beta.11 compiles against Amuru 2.0.0.

ganda repo audit first reported one warning, vscode-window-icon, because .vscode/settings.json was missing. ganda repo audit --fix --checks vscode-window-icon wrote that file. It is gitignored (.gitignore line .vscode/*). The re-audit reported Passed 38, Failed 0, Skipped 3.

On the clean task branch, ./bin/dev release --dry-run passed the gh and clean-tree guards and stopped with "release must be cut from master". git tag -l 'v1.1.0' was empty. ./bin/dev workflow --mode pr then ran self-install, clean, build, test, and pack and printed Pipeline SUCCEEDED. The packed file is artifacts/packages/TimeWarp.Builder.1.1.0.nupkg.

Review round 1's open findings were both in documentation/developer/guides/releasing.md. The guide says dev release walks back only over kanban/ (at most 50 commits). A docs-only HEAD is recovered with gh workflow run workflow.yml --ref master -f mode=merge, then another dry-run after that run is green. Download and Break-glass state that gh run rerun lasts 30 days. Past that, while master is still the tagged commit, the recovery is that merge-mode dispatch (the upload step requires refs/heads/master) and then break-glass on the tag. Round 1 merged.md marks M1 and M2 fixed. Round 2 verified those two and opened M3.

This fix pass reran ganda repo audit (Passed 38, Failed 0, Skipped 3), ./bin/dev build (0 warnings, both target DLLs), ./bin/dev test (6 of 6 Jaribu files), and ./bin/dev check-version (source 1.1.0, NuGet 1.0.0). ./bin/dev release --dry-run stopped at the clean-tree guard while these edits were uncommitted. After that commit, the same command passed the clean-tree guard and stopped with "release must be cut from master". It did not run gh release create. git tag -l 'v1.1.0' was empty. source/Directory.Build.props was unchanged.

Review round 2's bug was the pre-cut gate. Step 2 no longer reads the run id from guard 8. Guard 8 prints OrderCandidateRuns(...)[0], which is the push run when one exists, so a refresh leaves that line on the expired version-bump run. The guide now tells the operator to take the full SHA from the dry-run Target commit: line, list gh run list --workflow workflow.yml --commit SHA --status success, skip pull_request, and continue when any other run has a Packages-* artifact with expired false. Refresh is the same merge-mode dispatch, and only when every such artifact is missing or expired. The push run staying older than 7 days is not another refresh. Download says an expired push run is skipped so the later dispatch's artifact is the one promoted. Round 2 merged.md marks M3 fixed. Disposition waits for review round 3.

Review round 3's bug was the refresh loop that the M3 wording added. The merge-mode dispatch builds master HEAD. When HEAD is kanban-only commits past the dry-run target, the live Packages-* artifact is at HEAD, and the next dry-run moves Target commit: there. The guide no longer says to repeat the list at the SHA from before the refresh. After the run is green, the operator dry-runs again and lists at the new Target commit: SHA. That SHA stays when HEAD was already the target, and only then does guard 8 still name the old push run. It moves to HEAD when HEAD was kanban-only past the old target. Download no longer says the dispatch is "at the same commit". It says a later merge-mode dispatch on master supplies the bytes when that dispatch ran at the commit being released. Round 3 merged.md marks M4 fixed. Disposition waits for review round 4. No tag and no GitHub Release were created. <Version> is still 1.1.0.

This fix pass reran ganda repo audit (Passed 38, Failed 0, Skipped 3), ./bin/dev build (0 warnings, both target DLLs, package TimeWarp.Builder.1.1.0.nupkg), ./bin/dev test (6 of 6 Jaribu files), and ./bin/dev check-version (source 1.1.0, NuGet 1.0.0). ./bin/dev release --dry-run stopped at the clean-tree guard while these edits were uncommitted. After that commit, the same command passed the clean-tree guard and stopped with "release must be cut from master". It did not run gh release create. git tag -l 'v1.1.0' was empty. source/Directory.Build.props was unchanged.

Review round 4's bug was the guard 8 refusal paragraph. The walk names the first commit with a path outside kanban/. That commit is HEAD, or the commit under a run of kanban-only commits. On this repo the usual documentation-only merge is followed by land kanban commits, so the refusal names the merge. The clause "when HEAD is not kanban-only" is gone. The merge-mode dispatch builds HEAD, so after it is green the new Target commit: is HEAD. The operator dry-runs again and applies the artifact check at that line. Round 4 merged.md marks M5 fixed. Disposition waits for review round 5. No tag and no GitHub Release were created. <Version> is still 1.1.0.

This fix pass reran ganda repo audit (Passed 38, Failed 0, Skipped 3), ./bin/dev build (0 warnings, both target DLLs, package TimeWarp.Builder.1.1.0.nupkg), ./bin/dev test (6 of 6 Jaribu files), and ./bin/dev check-version (source 1.1.0, NuGet 1.0.0). ./bin/dev release --dry-run stopped at the clean-tree guard while these edits were uncommitted. After that commit, the same command passed the clean-tree guard and stopped with "release must be cut from master". It did not run gh release create. git tag -l 'v1.1.0' was empty. source/Directory.Build.props was unchanged.

Test plan

Smoke: from this worktree, on a clean tree:

./bin/dev --help
./bin/dev release --dry-run
./bin/dev check-version
./bin/dev build
./bin/dev test
git tag -l 'v1.1.0'
git diff -- source/Directory.Build.props
grep -n -E 'kanban-only|mode=merge|30 days|Target commit:|pull_request|OrderCandidateRuns' documentation/developer/guides/releasing.md

Expect: release is in ./bin/dev --help. ./bin/dev release --dry-run exits non-zero because this branch is not master, prints no gh release create execution, and git tag -l 'v1.1.0' stays empty. ./bin/dev check-version prints Packages checked: TimeWarp.Builder, Version in source: 1.1.0, and Latest NuGet version: 1.0.0. ./bin/dev build is 0 warnings and writes source/timewarp-builder/bin/Release/net10.0/TimeWarp.Builder.dll and source/timewarp-builder/bin/Release/net11.0/TimeWarp.Builder.dll. All 6 Jaribu runfiles pass. source/Directory.Build.props still has <Version>1.1.0</Version>. The guide's walk-back sentence names kanban-only commits and says a path outside kanban/, including documentation/, ends the walk. It contains gh workflow run workflow.yml --ref master -f mode=merge as the guard 8 recovery and as the pre-cut refresh. The pre-cut check copies the full SHA from Target commit:, runs gh run list --workflow workflow.yml --commit SHA --status success, skips pull_request, and continues when any other run has a Packages-* artifact with expired false. After the refresh is green it says to dry-run again and list at the new Target commit: SHA, and not to repeat the list at the SHA from before the refresh. It says that SHA moves to HEAD when HEAD was kanban-only commits past the old target, and that the old SHA still showing every artifact expired is not a reason to dispatch again. It limits "the push run that guard 8 names stays older than 7 days" to the case where HEAD was already the target. Download says a later merge-mode dispatch on master supplies the bytes when that dispatch ran at the commit being released. It says GitHub accepts gh run rerun for 30 days, and the break-glass section then dispatches merge mode on master while master is still the tagged commit. The guard 8 refusal starts with "no successful CI run of workflow.yml". When the walk reaches a commit with a path outside kanban/, it names that commit and the path. That commit is HEAD, or the commit under a run of kanban-only commits. The usual case is a documentation-only merge followed by land kanban commits, and the refusal names the merge. The dispatch builds HEAD, so after the run is green the new Target commit: is HEAD. The paragraph does not say the refused commit is HEAD, and it does not say "when HEAD is not kanban-only".

Automated gate: ./bin/dev build and ./bin/dev test from the repo root. ganda repo audit exits 0.

The release cut itself is not part of this validation. Run dev release only with --dry-run on this branch.

Kanban task 010
Session: 01a1bcc1-3c09-78e2-8fa0-ceeacadc6e70