---

name: agent-framework-py-release description: Use when cutting a Python release for the microsoft/agent-framework monorepo. Triggers on "bump py versions", "cut a python release", "prepare release PR for python", "release py packages", "bump python to X.Y.Z", or similar requests to bump Python package versions and prepare a release PR. Handles all four lifecycle tiers (alpha, beta, rc, released) with CHANGELOG-driven selective bumps, floor bound checks, and post-bump validation.

Agent Framework Python Release

Cuts a Python release PR for the microsoft/agent-framework monorepo.

Core principle: CHANGELOG drives bumps

No more universal lockstep. A package only bumps if it has a CHANGELOG entry this cycle. Draft the CHANGELOG first, then derive the bump list from it.

Exceptions to selective bumping (these stay coupled regardless of per-package changes):

• Root agent-framework is coupled to agent-framework-core via agent-framework-core[all]==X.Y.Z (exact pin in python/pyproject.toml). If core bumps, root bumps. Root may also bump independently for its own changes (docs/metadata/extras).
• Beta-tier cohort bumps are still allowed when the user explicitly wants a fresh date stamp across all betas for cohort signaling — but this is a deliberate user decision per release, not a default.

If the user explicitly states which packages to bump (or hands over a PR list per package), treat that as authoritative.

Lifecycle tiers and version rules

Use python/.github/skills/python-package-management/SKILL.md as the source of truth for lifecycle version patterns, date-stamp cutoffs, classifier alignment, and internal dependency updates.

For release work, derive the live tier map at release time from python/PACKAGE_STATUS.md and the actual version = lines in each pyproject.toml. Do not hardcode counts — packages move between tiers.

Inputs to confirm before bumping

1. The changeset: explicit commits/PRs the release covers, OR derive from git log ${LAST_RELEASED_TAG}..${RELEASE_BASE} -- python/.
2. Per-package CHANGELOG entries: which packages will get a line in the new release section. This list IS the bump list.
3. Per-released-package semver bump: for each released-tier package that has a CHANGELOG entry, decide PATCH / MINOR / MAJOR.
4. Date stamp (only if any alpha/beta is being bumped): default from the python-package-management date-stamp rule.
5. Optional cohort bump?: ask if betas should all get a new date stamp regardless of per-package changes. Default: no.

If the user states target versions or a date explicitly, use exactly what they said — do not "correct" based on historical timezones.

Non-negotiable rules

• Release workflow owns the CHANGELOG: individual feature/fix PRs do not edit python/CHANGELOG.md; this workflow creates the entries centrally from merged changes.
• CHANGELOG-driven bumps: only packages mentioned in the new CHANGELOG section get version bumps. Exceptions: root follows core (==pin); user-opted cohort bump on betas.
• Follow python-package-management for package lifecycle and versioning rules — do not duplicate those rules in this release workflow.
• No Co-Authored-By trailer on any commit.
• Use uv run for cross-version Python/Poe commands and add the relevant tooling/requirements-*.txt bundle only for single-version tools.
• Never rename an existing CHANGELOG section header during a new release cut. Only INSERT a new section above existing ones.
• Footer reference links are part of the CHANGELOG edit, not optional.

Workflow

1. Orient and create branch

git fetch origin main --tags --quiet
git fetch upstream main --tags --quiet 2>/dev/null || true
git status

# Fork clones use upstream/main as the authoritative release base; direct clones use origin/main.
if git show-ref --verify --quiet refs/remotes/upstream/main; then
  RELEASE_BASE=upstream/main
else
  RELEASE_BASE=origin/main
fi
git log -1 --oneline "$RELEASE_BASE"

If the user already has a bump-py-ver-release-* branch checked out, use it. Otherwise:

git checkout -b bump-py-ver-release-YYMMDD "$RELEASE_BASE"

2. Build the live tier map

Read python/PACKAGE_STATUS.md to enumerate current packages and lifecycle stages, and grep '^version' python/pyproject.toml python/packages/*/pyproject.toml to confirm actual versions. Cross-reference — if PACKAGE_STATUS.md shows a package as beta but its version looks like 1.0.0aYYMMDD, surface the inconsistency before proceeding.

Watch for: new packages added since the last release, lifecycle transitions (alpha→beta, beta→rc, rc→released), and stale packages that haven't been touched in many cycles.

3. Identify changeset

Use the LATEST released tag as the compare base:

LAST_RELEASED_TAG=$(
  git tag -l 'python-[0-9]*.[0-9]*.[0-9]*' \
    | uv run --directory python python -c 'import sys; tags=[t.strip() for t in sys.stdin if t.strip()]; print(max(tags, key=lambda t: tuple(map(int, t[7:].split(".")))))'
)
echo "Compare base: $LAST_RELEASED_TAG"

List commits and packages touched:

git log --oneline ${LAST_RELEASED_TAG}..${RELEASE_BASE} -- python/ ':!python/CHANGELOG.md'

# Per-commit package footprint
for sha in $(git log --format='%H' ${LAST_RELEASED_TAG}..${RELEASE_BASE} -- python/); do
  echo "--- $(git show -s --format='%h %s' $sha) ---"
  git show --name-only --format='' $sha | grep '^python/packages/' | \
    sed 's|^python/packages/||' | awk -F/ '{print $1}' | sort -u
done

When both remotes exist, record whether the fork is behind the authoritative base:

git rev-list --left-right --count origin/main...upstream/main

If user provides an explicit commit/PR list, treat THAT as authoritative.

3a. Build the authoritative touched-package set

Aggregate the per-commit footprint into a single union across the whole range. This is the source of truth for what must appear in the CHANGELOG. Save it before drafting entries.

# Union of all touched package directories across the range
git log --name-only --format='' ${LAST_RELEASED_TAG}..${RELEASE_BASE} -- python/packages/ \
  | grep '^python/packages/' \
  | sed 's|^python/packages/||' \
  | awk -F/ '{print $1}' \
  | sort -u

# Root-level files (drive a root agent-framework entry if substantive)
git log --name-only --format='' ${LAST_RELEASED_TAG}..${RELEASE_BASE} \
  -- python/pyproject.toml python/agent_framework_meta/ python/README.md \
  2>/dev/null | grep -v '^#x27; | sort -u

What "touched" means for bump purposes (apply judgment, document in commit message):

Change scope under python/packages/<pkg>/	Drives a bump?
Source code under <pkg>/agent_framework_* or the package's library tree	Yes — ship-affecting
pyproject.toml (deps, classifiers, extras, version metadata)	Yes — publishes new metadata
README.md substantive changes (env vars, install instructions, usage)	Yes — user-facing
README.md typo/wording-only	Judgment — usually no, but record under ### Changed if you do bump
Tests only (tests/)	Usually no — test changes don't ship to PyPI consumers. Bump only if the test change reflects a real behavior change in the package
Package-local samples (alpha packages only, before promotion)	Yes for alpha; samples ship with the alpha package
Repo infra touching the package dir (CI config, lint config)	No — record under - **tests**:, - **samples**:, or - **docs**: instead

Root agent-framework is touched when python/pyproject.toml, python/agent_framework_meta/, or root README.md substantive content changed.

4. Draft CHANGELOG entries (THIS DRIVES THE BUMP LIST)

Individual feature/fix PRs intentionally do not add CHANGELOG entries. During release preparation, derive this section centrally from the merged PRs since the last released tag.

Locate ## [Unreleased] and the top existing release header. INSERT a new section between them.

New section structure:

## [<release-label>] - YYYY-MM-DD

### Added
- **agent-framework-<pkg>**: <subject> ([#NNNN](https://github.com/microsoft/agent-framework/pull/NNNN))

### Changed
- **agent-framework-<pkg>**: ...

### Fixed
- **agent-framework-<pkg>**: ...

<release-label> is the released-tier target if any released package is bumping (e.g. 1.7.0), or the date stamp (e.g. 1.0.0b260528) if the release is prerelease-only. Reflect the user's framing.

Categorization heuristics (read commit subject + PR title/body):

• Added: new public APIs, new packages, new samples, feat(...), new re-exports from agent_framework, new capabilities surfaced from underlying SDKs
• Changed: metadata/classifier corrections, dependency upgrades, test infra changes, signature adjustments on existing APIs, renames, doc updates, lifecycle transitions
• Fixed: fix(...), bug/crash/regression repairs, noise reduction (e.g. stop emitting spurious warnings), protocol-compliance fixes
• Removed: deprecations/deletions (use [BREAKING] prefix if breaking)

Entry format:

• - **agent-framework-<pkg>**: <subject, grammar-normalized> ([#NNNN](https://github.com/microsoft/agent-framework/pull/NNNN))
• Multiple packages touched by one PR: comma-separated bold names at the head: - **agent-framework-core**, **agent-framework-foundry**: ...
• Root package changes: use - **agent-framework**: ...
• Test/sample/repo infra: - **tests**: ..., - **samples**: ..., - **docs**: ... (these do not drive package bumps)

Once entries are drafted, the set of **agent-framework-<pkg>** and root **agent-framework** mentions IS the bump list. Anything not mentioned does not bump.

4a. Reconcile mentions against the touched-package set (DO NOT SKIP)

Before moving on, prove that every ship-affecting touched package has at least one CHANGELOG entry. A package whose code changed but is missing from CHANGELOG will NOT bump — and the fix will not ship to PyPI consumers.

# 1. Touched ship-affecting packages and root package files (from step 3a)
TOUCHED_PACKAGES=$(git log --name-only --format='' ${LAST_RELEASED_TAG}..${RELEASE_BASE} -- python/packages/ \
  | grep '^python/packages/' \
  | sed 's|^python/packages/||' \
  | awk -F/ '{print $1}' \
  | sort -u)

ROOT_TOUCHED=$(git log --name-only --format='' ${LAST_RELEASED_TAG}..${RELEASE_BASE} \
  -- python/pyproject.toml python/agent_framework_meta/ python/README.md \
  2>/dev/null | grep -v '^#x27; | sort -u)

TOUCHED=$(
  {
    if [ -n "$TOUCHED_PACKAGES" ]; then echo "$TOUCHED_PACKAGES"; fi
    if [ -n "$ROOT_TOUCHED" ]; then echo "agent-framework"; fi
  } | sort -u
)

# 2. Packages mentioned in the new CHANGELOG section
# (Extract from the section you just drafted — adjust the awk range to the new section's bounds)
MENTIONED=$(awk '
  /^## \[<release-label>\]/ { in_section=1; next }
  in_section && /^## \[/ { exit }
  in_section { print }
' python/CHANGELOG.md \
  | grep -oE '\*\*agent-framework(-[a-z0-9_-]+)?\*\*' \
  | sed 's/\*\*//g;s/^agent-framework-//' \
  | sort -u)

# 3. Diff — anything in TOUCHED but missing from MENTIONED is a gap
comm -23 <(echo "$TOUCHED") <(echo "$MENTIONED")

For every gap, decide explicitly with the user:

• Add a CHANGELOG entry and bump — the default. Even small package changes (a dep upgrade, a metadata fix) deserve an entry under ### Changed.
• Intentionally skip — only when the touched files are demonstrably non-shipping (e.g. comments-only, test infra). Note the skip in the commit message body so reviewers see the reasoning.

Package-name aliasing to watch for: directory foundry_local → package agent-framework-foundry-local; github_copilot → agent-framework-github-copilot; azure-ai-search → agent-framework-azure-ai-search. The directory name and the published PyPI name are not always identical — confirm both when reconciling.

Footer reference links (REQUIRED every release):

grep -n "^\[.*\]:" python/CHANGELOG.md | head -5

Two edits when a released-tier bump is in play:

1. Advance [Unreleased] compare base from python-${OLD_RELEASED}...HEAD to python-${NEW_RELEASED}...HEAD.
2. INSERT a new [${NEW_RELEASED}] line ABOVE the previous version's link: [${NEW_RELEASED}]: https://github.com/microsoft/agent-framework/compare/python-${OLD_RELEASED}...python-${NEW_RELEASED}

For prerelease-only releases (no released-tier bump this cycle), footer links don't change.

5. Apply bumps per tier

For each package in the bump list, choose the rule for its current tier. Use anchored sed (^version = "..."$) so only the project's own version matches. Use sed -i.bak plus backup cleanup for portable in-place edits across macOS/BSD sed and GNU sed.

Released tier (per-package semver):

# Example: openai goes 1.6.0 -> 1.6.1 (PATCH); core stays
sed -i.bak 's/^version = "1.6.0"$/version = "1.6.1"/' python/packages/openai/pyproject.toml
rm python/packages/openai/pyproject.toml.bak

Root agent-framework (only when core bumps, OR for its own changes):

sed -i.bak "s/^version = \"${OLD_ROOT}\"$/version = \"${NEW_ROOT}\"/" python/pyproject.toml
# AND keep the exact pin in sync with core
sed -i.bak "s/agent-framework-core\\[all\\]==${OLD_CORE}/agent-framework-core[all]==${NEW_CORE}/" python/pyproject.toml
rm python/pyproject.toml.bak

RC tier (counter increment):

# ag-ui goes 1.0.0rc3 -> 1.0.0rc4 only because it has a CHANGELOG entry
sed -i.bak 's/^version = "1.0.0rc3"$/version = "1.0.0rc4"/' python/packages/ag-ui/pyproject.toml
rm python/packages/ag-ui/pyproject.toml.bak

Alpha/Beta tier (new date stamp):

OLD_DATE=260521
NEW_DATE=260528

# Only the packages with CHANGELOG entries get the new stamp.
# Example: anthropic and bedrock had entries, others did not.
for pkg in anthropic bedrock; do
  file="python/packages/$pkg/pyproject.toml"
  sed -i.bak "s/1\\.0\\.0b${OLD_DATE}/1.0.0b${NEW_DATE}/g" "$file"
  rm "${file}.bak"
done

If the user opts into a cohort-wide beta bump regardless of per-package changes, enumerate every beta package from PACKAGE_STATUS.md and stamp all of them. State that this is a cohort bump in the commit message.

Spot-check with grep '^version' python/pyproject.toml python/packages/*/pyproject.toml | sort before moving on.

5a. Refresh the core[all] release cohort

When Core bumps, update every Agent Framework dependency in python/packages/core/pyproject.toml [project.optional-dependencies].all:

• Workspace packages use their post-bump version as the lower bound, or their unchanged current version when they are not otherwise releasing this cycle.
• External Agent Framework packages such as Azure Functions and Durable Task retain a bounded known-good floor.
• Every entry keeps an upper bound for its supported major version.
• Preserve environment markers for packages that do not support every Python/platform combination.

These bounds prevent pip from backtracking across old cyclic core[all] -> integration -> core combinations. The release validator rejects missing, stale, or unbounded cohort entries.

6. Floor bound updates on agent-framework-core

Only relevant when core itself bumped this cycle. Two policies, pick one explicitly with the user:

• Conservative (default): raise agent-framework-core>=X.Y.Z to the new core version on every non-core package that is ALSO bumping this cycle. Leaves packages-not-bumped at their existing floor.
• Strict per-upstream-doc: only raise the floor on packages that actually consume a new core API introduced in the bump. This requires per-package code inspection because release probes use the co-released local core and cannot prove compatibility with an older published core floor.

When raising a core floor, replace only the >=OLD half of the bound you intend to change:

file="python/packages/<pkg>/pyproject.toml"
sed -i.bak "s/agent-framework-core>=${OLD_CORE}/agent-framework-core>=${NEW_CORE}/" "$file"
rm "${file}.bak"

If core did not bump this cycle, do not touch floors.

7. Validate

cd python && uv run poe validate-python-release --base-ref "$RELEASE_BASE"

Use the same freshly fetched main ref that the release branch was based on (upstream/main above; use origin/main when that is the authoritative release base). Must exit 0. This task first regenerates uv.lock, then discovers the package pyproject.toml files changed from that base and runs their published runtime dependencies and non-development extras through lock-independent lowest-direct and highest import probes. The probes run in parallel, derive the minimum supported Python minor from each package's internal editable closure, and share a hard 300-second deadline. Use --python only when the release requires an explicit interpreter override.

This is the release safety net for selective bumping: the lower probe catches unresolvable or unimportable external floors, internal constraints that reject co-released package versions, and the upper probe catches caps that exclude an installable package set. The JSON report records the concrete versions resolved in both scenarios. It does not replace the package-by-package code inspection required by the strict core-floor policy. If it fails, fix the named package/bound and re-run before committing.

Do not substitute the workspace-wide validate-dependency-bounds-test command here. That command runs every package's full tests and Pyright in separate isolated environments and is intentionally reserved for CI or an explicit dependency-range audit. If the release itself changes an external dependency range, also run validate-dependency-bounds-project --mode both --package <pkg> --dependency <name> for that dependency.

If only prereleases changed (no core bump, no floor changes), release validation is still required because the lockfile and both ends of each changed package's published dependency metadata must remain installable.

8. Commit (expect hook retry)

The project's pre-commit hook runs poe check / poe typing / uv lock and may re-modify uv.lock on first commit attempt. Expected pattern — run twice:

git add -u
git commit -m "<message>"   # may fail with "files were modified by this hook"
git add -u                  # re-stage hook-generated changes
git commit -m "<message>"   # succeeds

Commit message format — HEREDOC, no Co-Authored-By:

git commit -m "$(cat <<'EOF'
Bump Python package versions for <release-label> release

<One paragraph explaining:
 - which packages bumped and why (CHANGELOG-driven)
 - semver rationale for any released-tier bumps
 - whether a beta cohort bump was applied
 - whether core floors were raised, and under which policy>
EOF
)"

9. Push and report

git push origin bump-py-ver-release-YYMMDD

The push output includes a Create a pull request for '<branch>' on GitHub by visiting: <URL> line — surface that to the user.

Pitfalls from past cycles — avoid repeating

• Scoping from the wrong tag. A release may have been tagged between bump cycles. Always compute LAST_RELEASED_TAG fresh — do NOT reuse the previous compare base. Symptom: dragging already-shipped PRs into the new CHANGELOG section.
• Bumping packages that had no changes. This was the old lockstep default. Now: a package without a CHANGELOG entry does not bump. The two exceptions (root follows core, optional beta cohort bump) are explicit choices, not defaults.
• Touched package missing from CHANGELOG. The inverse failure mode of the previous pitfall, and the dangerous one. If a PR landed code into python/packages/foo/ but the new CHANGELOG section has no **agent-framework-foo**: line, then foo will NOT bump, and the fix sits in main but never reaches PyPI. Always run the touched-vs-mentioned reconciliation in step 4a before applying bumps.
• Directory name vs PyPI name confusion. Some package dirs use underscores or split names that differ from the published PyPI name (e.g. foundry_local → agent-framework-foundry-local). When reconciling touched dirs against CHANGELOG mentions, normalize on the PyPI form.
• Forgetting root when core bumps. Root agent-framework has agent-framework-core[all]==EXACT_VERSION. If core bumps and root does not, the pin is broken. Always bump them together and update the pin string.
• Renaming an existing section header. When drafting a new release, do NOT rewrite an existing ## [X.Y.Z] header to a new label — that wipes the historical section. Always INSERT a new section above.
• Forgetting footer reference links. When a released-tier bump is in play, the [Unreleased] line and the new [X.Y.Z] link at the bottom MUST be updated. Heading links don't resolve without them.
• Wrong footer compare base. The new [X.Y.Z] line compares from the PREVIOUS released tag, not from two releases ago.
• Timezone drift. For alpha/beta date stamps, follow the python-package-management date-stamp rule; do not infer a local timezone from the user's current shell.
• Co-Authored-By trailer. Never add it. Rewrite/amend if it slipped in.
• Stale inventory in this skill. Always read python/PACKAGE_STATUS.md for the live tier map. Do not trust a hardcoded list.
• Divergent origin vs upstream. In fork clones, use freshly fetched upstream/main consistently for branch creation, changeset discovery, and release validation. A stale origin/main must never become the implicit compare base.
• --pre README cleanup on promotion. When a package is promoted to released in this cycle, grep for pip install agent-framework-<pkg> --pre in READMEs and drop the --pre flag.
• RC counter inflation. Do not increment 1.0.0rcN without a CHANGELOG entry for that package. The counter tracks iterations, not calendar.

References

• Package lifecycle and versioning source of truth: python/.github/skills/python-package-management/SKILL.md
• Lifecycle source of truth: python/PACKAGE_STATUS.md
• Release validator: python/scripts/dependencies/validate_dependency_bounds.py --mode release (changed-package, lock-independent lowest-direct and highest import probes under a five-minute deadline)
• Poe task definitions: python/pyproject.toml [tool.poe.tasks]