Automate agentic draft release creation with reasoning-based notes generation

[Copilot]

This PR updates release automation so merged automated CMake-version PRs trigger an agentic, reasoning-based draft release flow instead of a deterministic direct publish flow.

• Workflow: agentic draft release creation

  • Updated .github/workflows/auto-release.yml to:
    • gather context (commits since previous release, recent merged PRs, previous release notes template)
    • run an LLM "Release Manager" reasoning step
    • decide whether a release draft is warranted
    • create a draft release via gh release create --draft only when approved by the reasoning output

• Reasoning output and release-note generation

  • The reasoning step produces structured output including:
    • release decision (should_create_draft_release)
    • release title
    • categorized changes (breaking/features/bug fixes/maintenance)
    • human-readable markdown release notes
  • Notes are generated from repository context and previous release style/template rather than simple tag string replacement.

• Resilience and fallback behavior

  • If models token is not configured or model request/parsing fails, workflow falls back to template-based draft creation.
  • Added robust parsing/error-handling improvements in the workflow scripting path.

• Docs: release process updated

  • Updated RELEASE_PROCESS.md to document:
    • agentic reasoning-based release preparation
    • draft release review/publish flow
    • distinction from manual tag-sync workflow behavior

- name: Create draft GitHub release
  run: |
    TAG_NAME="v${{ steps.check_version.outputs.CMAKE_VERSION }}"
    gh release create "$TAG_NAME" \
      --draft \
      --title "${{ steps.release_agent.outputs.RELEASE_TITLE }}" \
      --notes-file "${{ steps.release_agent.outputs.RELEASE_NOTES_FILE }}" \
      --verify-tag

[lukka]

I want an agentic, reasoning-based release process rather than a deterministic script, here is how you can set it up:

Trigger via GitHub Actions: Set up a workflow that runs on a specific trigger (e.g., a schedule, every push to main, or on pull request merges).
Gather Context: In the workflow step, use the GitHub CLI (gh) or REST API to gather recent commits, merged PRs, and relevant discussions.
Agentic Reasoning (LLM Step): Pass this context to an LLM (such as via GitHub Models, OpenAI API, or an AI action). You can prompt the model to act as a Release Manager to:
Determine if a release is actually warranted based on the changes.
Categorize the changes (breaking, features, bug fixes).
Generate intelligent, human-readable release notes based on the context.
Create the Draft Release: If the LLM determines a release should be made, the script parses the LLM's output and uses the GitHub CLI to create the draft release:
bash
gh release create --draft --title "" --notes ""
This approach allows a non-deterministic, context-aware agent to prepare the draft release for you, which a human can then review and publish.

[Copilot]

I want an agentic, reasoning-based release process rather than a deterministic script, here is how you can set it up:

Trigger via GitHub Actions: Set up a workflow that runs on a specific trigger (e....

Implemented in commit 410a1f3. The auto-release workflow now follows an agentic flow: it gathers release context (commits, merged PRs, previous release template), runs an LLM release-manager reasoning step, and conditionally creates a draft release via gh release create --draft with generated notes. It also includes robust fallback behavior when model token/call/parsing fails, and the release-process docs were updated to reflect review-and-publish of drafts.