Make GitHub release tags short (rust-<v> / js-<v>) and fix missing npm/crates.io/GitHub releases

[konard]

Summary

Closes #17. Makes GitHub release tags short — rust-<version> and js-<version> (no v prefix, single - separator) — and fixes the "I still don't see all the releases in package managers and GitHub" symptom by repairing three independent release-pipeline faults found while auditing the two runs the issue links to.

Full analysis: docs/case-studies/issue-17/README.md and template-comparison.md.

Release-channel state at investigation time

Channel	Published	On main	Diagnosis
npm	0.8.2	0.10.1	bumped four releases ahead but never published
JS GitHub Releases	none	—	no JS release ever created
crates.io	0.2.0	0.2.0	crate version never advanced (no bump step)
Rust GitHub Releases	0.2.0 ×2	—	duplicate rust-v / rust_v tags for one version

Root causes & fixes

• RC-1 — tag format (R1). getTagPrefix() → js- (multi) / '' (single); Rust workflow → rust-$VERSION / $VERSION. normalizeVersion() still strips every legacy spelling (v, js-v, js_v, rust-v, rust_v) so old tags remain valid inputs.
• RC-2 — JS releases stuck (R3). When version-and-commit.mjs finds nothing to commit (a prior run bumped + pushed but died before publishing, consuming the changeset), it now also sets already_released=true. The idempotent publish step then verifies npm and self-heals the missing version instead of skipping it forever. This is why npm sat at 0.8.2 while package.json was 0.10.1.
• RC-3 — Rust version never advances (R3). The Rust release job had no bump step, so the crate was pinned at 0.2.0 forever. Added get-bump-type.rs / bump-version.rs / collect-changelog.rs (mirroring the JS changeset flow and the link-foundation rust template) and wired them into the release job: re-align to origin/main (concurrency-safe, no double-bump), pick the highest declared bump, rewrite Cargo.toml, fold changelog.d/ fragments into CHANGELOG.md, cargo update -p the crate, commit rust-<version>, push, and target the release at the bump commit.
• RC-4 — Rust publish was a false positive (R2). The post-publish smoke test broke on a transitive alloc-no-stdlib 2.0.4/3.0.0 split (E0277) even though 0.2.0 was already published. The smoke test now collapses the duplicate major to 2.0.4 on the fresh consumer resolve (no-op once upstream realigns).
• RC-5 — duplicate Rust tags (R2). Tag convention centralised; the short scheme yields exactly one tag per version going forward. Existing duplicate 0.2.0 releases are left in place — deleting published releases is destructive/outward-facing, so the case study documents a manual cleanup recommendation instead.
• R4/R9 — best practices / Node 20 deprecation. Action pins bumped to Node 20+ runtimes (checkout@v6, setup-node@v6, cache@v5, create-pull-request@v8) across js.yml, rust.yml, parity.yml.

What happens on the first push to main after merge

1. Rust bumps 0.2.0 → 0.3.0, publishes to crates.io, creates release rust-0.3.0.
2. JS publishes 0.10.1 to npm (self-heal if still in limbo) and creates release js-0.10.1.

Tests / verification

• npm test: 158/158 pass (includes updated release-naming tag tests — buildReleaseTag('1.2.3', MULTI) === 'js-1.2.3', never contains v/_; normalizeVersion accepts all legacy spellings).
• npm run lint and npm run format:check: clean (format check covers .github/workflows).
• Rust scripts compile clean under RUSTFLAGS=-Dwarnings; exercised locally: 5 fragments → minor; 0.2.0 → 0.3.0; changelog folded; fragments consumed; Cargo.lock synced.
• All three workflow YAMLs parse.

Upstream reports (R5)

Three shared template defects were verified against the live JS template source and filed upstream (repro + fix in each issue):

• js-template#85 — extractReleaseNotes regex lacks an end-anchor: notes for 1.2 match the 1.2.3 section.
• js-template#86 — check-changesets counts any .md file as a changeset.
• js-template#87 — merge-changesets silently drops malformed changesets.

The full catalogue (including lower-severity items) is in template-comparison.md §"Shared bugs to report upstream".

Reproduction & case study (R6/R7)

docs/case-studies/issue-17/ holds the deep analysis, the raw failing/passing run logs, the issue/run/registry-state JSON snapshots, and the upstream reference scripts.

[konard]

Working session summary

All three workflows pass on the final commit; the tree is clean. Everything is complete. Here's the final state.

Summary

Issue #17 is fully solved in PR #18 (ready for review, CI green on all three workflows).

The real bug behind "I don't see all the releases"

Auditing the two linked runs revealed three independent pipeline faults, on top of the tag-format request:

Channel	Was	Root cause	Fix
npm 0.8.2 vs package.json 0.10.1	bumped but never published	a prior run consumed the changeset then died before publishing; next run found "nothing to commit" and the publish gate skipped forever	version-and-commit.mjs now signals already_released so the idempotent publish self-heals
crates.io stuck at 0.2.0	version never advanced	the Rust release job had no bump step	new get-bump-type.rs / bump-version.rs / collect-changelog.rs wired into rust.yml (mirrors the JS changeset flow + the rust template)
Rust publish job red	false positive	post-publish smoke test broke on a transitive alloc-no-stdlib 2.0.4/3.0.0 split (E0277) even though 0.2.0 was already published	smoke test collapses the duplicate major to 2.0.4

Tag format (the headline request)

• getTagPrefix() → js- (multi) / '' (single); Rust workflow → rust-$VERSION / $VERSION — short, no v, single -.
• normalizeVersion() still accepts every legacy spelling (v, js-v, js_v, rust-v, rust_v), so old tags remain valid inputs.
• After merge, the next push produces rust-0.3.0 and js-0.10.1.

Also done

• Action pins bumped to Node 20+ runtimes across all three workflows.
• Deep case study at docs/case-studies/issue-17/ (timeline, requirement matrix R1–R10, root causes RC-1–RC-5, libraries considered, raw logs + JSON data snapshots, full template comparison).
• Three upstream defects filed against the JS template after verifying them against live source: #85 (changelog regex prefix-collision), #86 (changeset .md over-counting), #87 (merge-changesets silent drop).

One thing I deliberately did not do

The duplicate rust-v0.2.0 / rust_v0.2.0 releases for the same version are left in place — deleting published releases is destructive and outward-facing, so the case study (§7) documents a manual gh release delete recommendation rather than executing it. The short-tag scheme yields exactly one tag per version going forward, so it can't recur.

Verification: npm test 158/158, npm run lint + format:check clean, Rust scripts compile clean under -Dwarnings and exercised locally (0.2.0→0.3.0, 5 fragments→minor), all YAMLs parse, CI green.

---

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 estimation:

• Public pricing estimate: $15.553973
• Calculated by Anthropic: $17.242778
• Difference: $1.688805 (+10.86%)

📊 Context and tokens usage:

Claude Opus 4.8: (4 sub-sessions)

1. 115.6K / 1M (12%) input tokens, 26.4K / 128K (21%) output tokens
2. 114.8K / 1M (11%) input tokens, 33.4K / 128K (26%) output tokens
3. 116.3K / 1M (12%) input tokens, 32.7K / 128K (26%) output tokens
4. 108.0K / 1M (11%) input tokens, 36.5K / 128K (29%) output tokens

Total: (72.1K new + 581.9K cache writes + 13.7M cache reads) input tokens, 176.6K output tokens, $15.278797 cost

Claude Haiku 4.5:

• 142.0K / 200K (71%) input tokens, 12.3K / 64K (19%) output tokens

Total: (89 new + 141.9K cache writes + 361.1K cache reads) input tokens, 12.3K output tokens, $0.275177 cost

🤖 Models used:

• Tool: Anthropic Claude Code
• Requested: opus
• Main model: Claude Opus 4.8 (claude-opus-4-8)
• Additional models:
  • Claude Haiku 4.5 (claude-haiku-4-5-20251001)

📎 Log file uploaded as Gist (6091KB)

• View complete solution draft log

---

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

[konard]

✅ Ready to merge

This pull request is now ready to be merged:

• All CI checks have passed
• No merge conflicts
• No pending changes

---

Monitored by hive-mind with --auto-restart-until-mergeable flag