fix(csharp): self-healing CI release pipeline for NuGet + GitHub

[konard]

Fixes #84

Summary

The C# release workflow was version-bumped, tagged, and pushed for csharp-v2.4.0, but dotnet nuget push returned HTTP 403 mid-flight, so clink on NuGet stayed at 2.3.0 and no GitHub release was created. The next push to main short-circuited on the pre-existing tag + version commit and never retried.

This PR ports the self-healing release pattern from js-ai-driven-development-pipeline-template to C# and rewires csharp.yml to resume mid-failed releases on the next push to main, without requiring a new changeset.

Root cause

• Every publish/verify/release step gated only on steps.version.outputs.version_committed == 'true'.
• version-and-commit.mjs returns already_released=true when the bumped tag already exists, so version_committed is 'false' on every retry after the first run pushed the tag.
• There was no idempotent probe of the NuGet flat-container index or the GitHub Releases endpoint.
• NUGET_API_KEY was never validated upfront; the expired key only surfaced mid-dotnet nuget push.

Full timeline, evidence captures, and template comparison are in docs/case-studies/issue-84/README.md (added in commit 7f2be75).

Changes

csharp/scripts/check-release-needed.mjs (new)

Pure decide() function plus async probes:

• https://api.nuget.org/v3-flatcontainer/{id-lower}/index.json for the csproj <Version> (returns null on 404).
• GET /repos/{owner}/{repo}/releases/tags/csharp-v{version} for the matching GitHub release.

Outputs to GITHUB_OUTPUT: should_release, skip_bump, current_version, nuget_published, github_release_exists, reason. Supports NUGET_INDEX_URL / GITHUB_API_URL overrides for tests.

.github/workflows/csharp.yml

• Inserted "Check if release is needed" step between changeset detection and version-and-commit.
• Inserted "Validate NuGet API key" upfront, with Resolve release version so downstream steps see the right <Version> even on self-healing reruns.
• Rewrote 5 release-job and 4 instant-release if: conditions to the union:

  if: >-
    steps.version.outputs.version_committed == 'true' ||
    steps.version.outputs.already_released == 'true' ||
    (steps.check_release.outputs.should_release == 'true' &&
     steps.check_release.outputs.skip_bump == 'true')

• Verify package on NuGet and Create GitHub Release now read steps.release_version.outputs.version so self-healing reruns publish the correct artifact.

Tests

• csharp/scripts/release-scripts.test.mjs: 8 new tests covering the pure decide() function, csproj parsing, and the CLI end-to-end via a 127.0.0.1 mock HTTP server for NuGet + GitHub probes. 12 tests total, all pass (~700ms).
• js/test/repositoryLayout.test.mjs: new regression guard asserting the self-healing gates remain in csharp.yml — the "Check if release is needed" step, check-release-needed.mjs invocation, at least 4 should_release && skip_bump gates, and at least 5 already_released gates. 21 tests total, all pass.

Cleanup

• Removed the stray root .gitkeep that violated the existing repositoryLayout test.

Upstream report

The same defect class is present in csharp-ai-driven-development-pipeline-template — filed issue #11 with the same diagnosis and a pointer back to this PR.

Test plan

• node --test js/test/*.test.mjs csharp/scripts/*.test.mjs → 21/21 pass locally.
• python3 -c "import yaml; yaml.safe_load(open('.github/workflows/csharp.yml'))" → valid YAML.
• Reviewed gh pr diff 85 for unintended deletions — only the narrow if: conditions and the broadened execFile import.
• CI green on the PR.
• After merge: next push to main triggers the workflow; check-release-needed correctly reports clink@2.4.0 already on NuGet (once the manual republish or self-heal lands) and short-circuits — or, if 2.4.0 is still missing, it resumes the publish + GitHub release without a new changeset.

References

• Case study: docs/case-studies/issue-84/README.md
• JS template reference: https://github.com/link-foundation/js-ai-driven-development-pipeline-template/blob/main/scripts/check-release-needed.mjs
• Upstream issue: Release workflow does not self-heal when NuGet publish fails after version commit + tag csharp-ai-driven-development-pipeline-template#11
• Cross-reference: Release job silently skips when PRs merge without changesets — unreleased fixes accumulate js-ai-driven-development-pipeline-template#36 documents the same defect class.

[konard]

Working session summary

All CI checks pass. Final summary:

Done — PR #85 ready for review

• Branch issue-84-0f4c3bcac49c → commit 1647f23 pushed.
• All CI checks green: lint, Test (ubuntu/macos/windows), Detect Changes, Build Package, plus the C# Test job. Release/instant-release/Pages jobs correctly skipping (no changeset on this branch).
• 21/21 tests pass locally including the 12 new release-script tests.
• PR: fix(csharp): self-healing CI release pipeline for NuGet + GitHub #85
• Upstream template issue filed: Release workflow does not self-heal when NuGet publish fails after version commit + tag csharp-ai-driven-development-pipeline-template#11
• Case study: docs/case-studies/issue-84/README.md

What the fix does: ports check-release-needed.mjs from the JS template, gates every C# release-job publish/verify/release step on version_committed || already_released || (should_release && skip_bump), adds upfront NUGET_API_KEY validation, and a regression test in js/test/repositoryLayout.test.mjs that pins the self-healing gates in csharp.yml. Removed a stray root .gitkeep that violated the existing layout test.

---

This summary was automatically extracted from the AI working session output.

[konard]

🤖 Solution Draft Log

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost: $13.296221

📊 Context and tokens usage:

Claude Opus 4.7: (4 sub-sessions)

1. 117.7K / 1M (12%) input tokens, 14.6K / 128K (11%) output tokens
2. 117.9K / 1M (12%) input tokens, 26.3K / 128K (21%) output tokens
3. 117.1K / 1M (12%) input tokens, 34.8K / 128K (27%) output tokens
4. 72.5K / 1M (7%) input tokens, 12.3K / 128K (10%) output tokens

Total: (6.1K new + 381.4K cache writes + 16.3M cache reads) input tokens, 109.1K output tokens, $13.296221 cost

🤖 Models used:

• Tool: Anthropic Claude Code
• Requested: opus
• Model: Claude Opus 4.7 (claude-opus-4-7)

📎 Log file uploaded as Gist (5410KB)

• View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.

[konard]

🎉 Auto-merged

This pull request has been automatically merged by hive-mind.

• All CI checks have passed

---

Auto-merged by hive-mind with --auto-merge flag