Scope PyPI release tags for the CLI package

- Change the PyPI release workflow to trigger only on pypi-glasskit-ai-v* tags so this monorepo can support other package release streams later.
- Validate the package version against the package-specific tag prefix and title GitHub Releases as PyPI: glasskit.ai vX.Y.Z.
- Replace GitHub repo-wide generated release notes with a small package-scoped note body and cli/ commit list since the previous PyPI package tag.
- Update CLI release docs and agent notes to use the new package-specific tag format.
- Verified with UV_PYTHON=3.12 uv run ty check, uv run pytest, uv run ruff check --fix, uv run ruff format, source CLI help checks, uv build --no-sources --clear, isolated wheel/sdist help smoke tests, a release-note shell simulation with pipefail, and git diff --check.

Author: tash-2s

.github/workflows/release.yml

@@ -6,7 +6,7 @@ env:
 on:
   push:
     tags:
-      - "v*"
+      - "pypi-glasskit-ai-v*"
 
 concurrency:
   group: release-${{ github.ref }}
@@ -55,7 +55,7 @@ jobs:
         shell: bash
         run: |
           PACKAGE_VERSION="$(uv version --short)"
-          TAG_VERSION="${GITHUB_REF_NAME#v}"
+          TAG_VERSION="${GITHUB_REF_NAME#pypi-glasskit-ai-v}"
 
           if [[ "${PACKAGE_VERSION}" != "${TAG_VERSION}" ]]; then
             echo "::error::pyproject.toml version ${PACKAGE_VERSION} does not match tag ${TAG_VERSION}"
@@ -94,9 +94,37 @@ jobs:
       - name: Create GitHub Release
         env:
           GH_TOKEN: ${{ github.token }}
+          RELEASE_TAG_PREFIX: pypi-glasskit-ai-v
         run: |
+          VERSION="${GITHUB_REF_NAME#${RELEASE_TAG_PREFIX}}"
+          PREVIOUS_TAG="$(
+            git tag --list "${RELEASE_TAG_PREFIX}*" --sort=-v:refname \
+              | grep -Fxv "${GITHUB_REF_NAME}" \
+              | head -n 1 \
+              || true
+          )"
+
+          {
+            echo "PyPI package: glasskit.ai"
+            echo "Version: ${VERSION}"
+
+            if [[ -n "${PREVIOUS_TAG}" ]]; then
+              echo
+              echo "CLI changes since ${PREVIOUS_TAG}:"
+              git log --format='- %s (%h)' "${PREVIOUS_TAG}..${GITHUB_REF_NAME}" -- cli > /tmp/cli-release-changes.txt
+              if [[ -s /tmp/cli-release-changes.txt ]]; then
+                cat /tmp/cli-release-changes.txt
+              else
+                echo "- No changes under cli/."
+              fi
+            else
+              echo
+              echo "Initial PyPI release for glasskit.ai."
+            fi
+          } > release-notes.md
+
           gh release create "${GITHUB_REF_NAME}" dist/* \
             --repo "${GITHUB_REPOSITORY}" \
             --verify-tag \
-            --title "${GITHUB_REF_NAME}" \
-            --generate-notes
+            --title "PyPI: glasskit.ai v${VERSION}" \
+            --notes-file release-notes.md

cli/AGENTS.md

@@ -30,5 +30,5 @@ GlassKit CLI turns recorded smart-glasses workflows into repeatable eval suites
 - `uv run ty check && uv run pytest && uv run ruff check --fix && uv run ruff format`: run after changes.
 - `uv run glasskit --help` and `uv run glasskit eval --help`: smoke-check the console entry point.
 - `uv build --no-sources --clear` plus isolated `dist/*.whl` and `dist/*.tar.gz` smoke tests: verify publishable artifacts before release.
-- Releases are tag-triggered from the repository root workflow. `pyproject.toml` version and the pushed `vX.Y.Z` tag must match.
+- PyPI releases are tag-triggered from the repository root workflow. `pyproject.toml` version and the pushed `pypi-glasskit-ai-vX.Y.Z` tag must match.
 - For local testing against the Origami backend, run the CLI from the app backend directory so local adapter imports resolve naturally: `cd REPO-ROOT/examples/origami/backend && uv run --with-editable ../../../cli --env-file .env glasskit eval run --adapter eval_adapter.py:create_evaluator --suite eval-suite`

cli/PUBLISHING.md

@@ -1,6 +1,6 @@
 # Publishing GlassKit CLI
 
-The PyPI project is `glasskit.ai`, and the installed command is `glasskit`. Releases are tag-driven from GitHub Actions using PyPI Trusted Publishing, so no PyPI token should be stored in GitHub.
+The PyPI project is `glasskit.ai`, and the installed command is `glasskit`. PyPI releases are tag-driven from GitHub Actions using PyPI Trusted Publishing, so no PyPI token should be stored in GitHub.
 
 ## One-Time Setup (Done)
 
@@ -33,18 +33,18 @@ uv run --isolated --no-project --with dist/*.whl glasskit --help
 uv run --isolated --no-project --with dist/*.tar.gz glasskit --help
 ```
 
-For a normal future release, bump the version, commit the version change, tag the commit, and push both the branch and tag:
+For a normal future release, bump the version, commit the version change, tag the commit with the PyPI package-specific tag format, and push both the branch and tag:
 
 ```bash
 uv version --bump patch
 VERSION="$(uv version --short)"
 git add pyproject.toml uv.lock
-git commit -m "Release v${VERSION}"
-git tag -a "v${VERSION}" -m "v${VERSION}"
+git commit -m "Release glasskit.ai v${VERSION}"
+git tag -a "pypi-glasskit-ai-v${VERSION}" -m "pypi-glasskit-ai-v${VERSION}"
 git push origin main
-git push origin "v${VERSION}"
+git push origin "pypi-glasskit-ai-v${VERSION}"
 ```
 
-Pushing the tag runs `.github/workflows/release.yml`. The workflow checks that the tag matches `pyproject.toml`, runs type checks, tests, lint, formatting checks, source CLI help checks, builds wheel and sdist artifacts, smoke-tests both artifacts, publishes to PyPI with `uv publish --trusted-publishing always`, and creates a GitHub Release with the built artifacts attached.
+Pushing the `pypi-glasskit-ai-vX.Y.Z` tag runs `.github/workflows/release.yml`. The workflow checks that the tag matches `pyproject.toml`, runs type checks, tests, lint, formatting checks, source CLI help checks, builds wheel and sdist artifacts, smoke-tests both artifacts, publishes to PyPI with `uv publish --trusted-publishing always`, and creates a package-scoped GitHub Release with the built artifacts attached.
 
 After pushing the release tag, open the GitHub Actions release run and approve the `pypi` deployment when GitHub asks for review.