✅ Done! Here's the PR and enyst's comments. The PR #18 â feat: add opt-in freeze-on-ready to release-ready.yml (closed) <https://github.com/OpenHands/release-actions/pull/18 by|github.com/OpenHands/release-actions/pull/18_by> vasco (VascoSch92) Vasco has 5 PRs total on

OpenHands/release-actions

(#12, #14, #17, #18, #20), but only #18 has comments from enyst. enyst's comments on PR #18 enyst left 6 comments, mostly reviewing the release workflow and asking vasco to refine the human/agent recipe: 1. Kicks off the discussion â asks to understand the move from the SDK's old

prepare-release

skill to the new release-please-based action, and to spell out the "Before/After" human and agent workflows. 2. Traces the workflow (agent-generated) â describes the full Before/After of the release flow: release-please keeps a draft

chore(main): release X.Y.Z

PR; marking it ready (with

freeze-on-ready: true

) cuts a

release/X.Y

branch and re-drafts the main PR; a frozen

chore(release/X.Y)

PR is then tested and merged. Flags that SDK PR #3939 still references the deleted

prepare-release.yml

skill. 3. Wording/procedure notes â suggests the frozen PR could be opened/marked ready automatically so the human doesn't act twice; asks about the

X.Y

vs

X.Y.Z

naming; wonders whether the agent will know which release branch to backport to. 4. Agent workflow tweaks â proposes the agent (not the human) marks PRs ready; asks whether to auto-switch the

chore(main)

PR back to draft; questions "guide the merge-back" (should the skill just do it?) and notes release-please already generates release notes. 5. Asks vasco directly â "I would really love it if we can give the human a recipe or two recipes⦠Maybe as commands?

/release

and `/release-merge`". Also suggests renaming "frozen" â "current" for less error-prone wording. 6. Clarifies the freeze rationale â confirms SDK releases are cut from main but stresses the test window: three (live LLM) test suites run during the release PR's life; if release-please keeps pulling in random main commits, all tests must always re-run, raising cost/timing/stability concerns â especially for small patches. Want me to pull anything else from that thread?