Clean up stale docs and add superseded banners to old PR write-ups

AGENTS.md

@@ -4,6 +4,10 @@
 
 Do NOT bump versions or publish locally. Use the GitHub Actions workflow:
 
-1. Go to Actions > Release > Run workflow
+1. Go to Actions > Publish > Run workflow
 2. Select version bump type (patch/minor/major)
-3. The workflow will bump version, push tags, create a GitHub release, and publish to npm automatically
+3. The workflow will run tests, bump version, push tags, create a GitHub release, and publish to npm automatically
+
+### Republishing a stalled version
+
+If a previous run tagged the version and created the GitHub release but failed at the npm publish step (rare, usually infrastructure-related), re-run the workflow with the `skip_bump` input set to true. This will republish the current `package.json` version without bumping or re-tagging.

documentation/PULL_REQUESTS/2026-05-01-cleanup-red-herrings.md

@@ -0,0 +1,19 @@
+## Problem
+
+After the debugging spiral that preceded the `v1.1.0` release, a review surfaced two places on `main` where the documentation no longer matches reality:
+
+1. Our release instructions still point to a workflow name that no longer exists (the file was renamed during the fix).
+2. Three of the PR write-ups from the debugging spiral describe fixes that didn't work or were later reverted, with no pointer to the PR that actually fixed the issue. A future reader finding them in isolation could easily be misled.
+
+## Solution
+
+Update the release instructions to match the current workflow name, and add a "Superseded" banner to each of the three misleading PR write-ups with a link to the real fix. Historical content is preserved in full.
+
+## Impact
+
+No behavior changes. Anyone reading `AGENTS.md` or the old PR write-ups now gets accurate guidance.
+
+# Credits
+
+- Nabs (Architect)
+- JENA (Lead Developer)

documentation/PULL_REQUESTS/2026-05-01-fix-release-corepack.md

@@ -1,3 +1,5 @@
+> **⚠️ Superseded:** The corepack-based approach described here was reverted. The real root cause was unrelated — the npm Trusted Publisher was registered against the old workflow filename `publish.yml`, but the workflow had been renamed to `release.yml`. See [`2026-05-01-fix-release-node20.md`](./2026-05-01-fix-release-node20.md) for the fix that actually worked. Preserved for historical context.
+
 ## Problem
 
 Our release pipeline is currently unable to publish new versions to npm. The step that upgrades the installed `npm` to a version new enough to use our secure authentication method fails with an internal error — not because the target version is wrong, but because the copy of `npm` preinstalled on our build server has corrupted dependencies and cannot install anything, including itself.

documentation/PULL_REQUESTS/2026-05-01-fix-release-npm-auth.md

@@ -1,3 +1,5 @@
+> **⚠️ Partially superseded:** The `skip_bump` input added here is still in use, but the `npm@11` version pin was later reverted to `npm@latest`. See [`2026-05-01-fix-release-node20.md`](./2026-05-01-fix-release-node20.md) for the complete fix. Preserved for historical context.
+
 ## Problem
 
 The previous release workflow attempt reached the publish step but failed to authenticate with npm, even though everything upstream had completed successfully — tag created, GitHub release page live. The underlying cause was that the `npm` version shipped with the GitHub Actions runner is too old to use our secure publishing path.

documentation/PULL_REQUESTS/2026-05-01-fix-release-npm-upgrade.md

@@ -1,3 +1,5 @@
+> **⚠️ Superseded:** This change was later reverted. The diagnosis here was incorrect — the `npm` upgrade step was actually necessary. See [`2026-05-01-fix-release-node20.md`](./2026-05-01-fix-release-node20.md) for the fix that actually worked. Preserved for historical context.
+
 ## Problem
 
 Our release workflow failed on the very first run after introducing CI — not on any of our changes, but on an infrastructure step that was already there: an attempt to upgrade npm that no longer works on the current GitHub Actions runner. Tests passed, so nothing broken was shipped, but the version bump and publish never happened.