Skip to content

docs: curate all 14 historical MCP Server release descriptions #529

Open

Description

@yeongseon

Goal

Review, correct, publish and independently verify the 14 existing published CUBRID MCP Server GitHub Release descriptions, with factually correct changes and human contributor attribution. Preserve each release's real MCP runtime/security contracts and Git tag history.

Verified inventory (2026-11-08)

The live GitHub Releases API returned 14 published, non-draft Releases: v0.1.0, v0.2.1, v0.3.0, v0.3.1, v0.4.0, v0.5.0, v0.5.1, v0.5.2, v0.5.4, v0.5.5, v0.5.6, v0.6.0, v0.7.0, v0.8.0. Do not invent missing v0.2.0 or v0.5.3 Releases. The candidate v0.8.1 remains an open PR #526, not a published version in this baseline.

The v0.5.0 body is 52,109 characters; v0.5.1/v0.5.2 are 16,406/17,120; v0.5.4 is 16,120. These need content-loss-aware review. A literal by @ scan is not a full author audit: v0.2.1 has credits, while most others have none in that syntax.

Missing-number and package-provenance reconciliation

Sub-issue execution — version ownership (parallel)

The following work packages have disjoint version-specific editing scope. They can be worked on concurrently; none should rewrite another package's CHANGELOG version sections or published bodies.

Coverage: 5 + 1 + 2 + 3 + 3 = exactly 14 versions, with no overlapping owner.

Independent audit and final publication evidence

  • docs: independently verify human contributor credits across MCP Server releases #539 — 14/14 human contributor and first-contribution audit. Independently verify tag-range commits, merged PR creators and co-authors. Report missing/incorrect/unverifiable credits without concurrently editing another child's source section. Humans using AI tools remain credited; bots/AI agents/release publisher accounts are not falsely labeled human contributors.
  • release: independently verify all 14 published MCP Server notes #540 — 14/14 final independent public-release audit, including approved canonical body comparison, SHA-256, fresh GitHub API reads, tag/date/metadata/assets/flags preservation, links, unique bullets and factual MCP permissions/migration caveats. Prepare baseline in parallel; final sign-off follows reviewed publication.

Per-release acceptance contract

  1. Snapshot original body, release ID, tag, date, flags, asset inventory, version-specific Full Changelog target and a body digest.
  2. Verify only this repository's actual tagged source, merged PR/commit authors and CHANGELOG.md. Do not import facts or human authors from pycubrid or sqlalchemy-cubrid.
  3. Preserve read-only defaults, SQL allow-list and opt-in write protections, credential/error redaction behavior, tool/resource/prompt surface, transport and FastMCP client requirements, integration verification scope and known migration risks as supported by the exact tag. Distinguish tested/claimed/unknown versions.
  4. Produce readable, semantically deduplicated, appropriately grouped Markdown. Verify issue/PR numbers, links, release-to-release comparisons, and human by @.../first-contribution labels. If condensing historically lengthy Releases, preserve detailed source facts in repository-local long-form history.
  5. After independently reviewed source edits, perform authorized GitHub Release-body-only synchronization protected by live-body drift checks, and re-fetch each published Release. A CHANGELOG merge or green CI is not equivalent to public publication. Preserve all tags, original dates, draft/prerelease/latest flags and assets; never republish or mutate PyPI distributions as part of this effort.
  6. Close this parent only after release: independently verify all 14 published MCP Server notes #540 records 14/14 public proof and exceptions are resolved. If a new normal release is published meanwhile, separately reconcile its scope; do not silently count it among the original 14.

Complementary local issues — do not duplicate

  • bug: List each change once in CHANGELOG and release notes #506 — prevent release-please/curated CHANGELOG duplicate entries in the composer. Historical body-specific corrections remain in the 5 version owners.
  • chore: audit and remove obsolete temporary code and tooling #534 — temporary script/workflow inventory and safe cleanup. Specifically do not remove .github/workflows/sync-historical-release-footers.yml or scripts/sync_historical_release_footers.py while required for the above historical publishing/verification, unless a reviewed and functioning replacement method is documented. Actual unused status must be verified.
  • docs: define when temporary tooling must be reviewed and removed #535 — lightweight temporary-tool lifecycle documentation, independent of editing Releases.
  • CI policy: scripts/ci_scope.py already distinguishes Markdown docs-only PRs (no live lane), while ordinary tooling changes run unit tests and CI-policy changes run conservative/full checks. Do not copy a release-tooling special CI lane from another repository without first demonstrating a missing, costly test-selection case. Preserve the default read-only live safety checks, main/release and scheduled integration matrices.
  • docs: add an MCP demo GIF and Claude Desktop walkthrough video #164 already owns MCP demo imagery; do not recreate it as a release-note subtask.

Definition of done

Repository ownership: this tracker is limited to cubrid-lab/cubrid-mcp-server. Cross-repository links are informational only, not scope or release-blocking dependencies.

Activity

  1. added
    documentationImprovements or additions to documentation
    and removed on Nov 8, 2026
  2. yeongseon commented on Nov 18, 2026

    @yeongseon
    ContributorAuthor

    2026-11-18 current public Release evidence

    Fresh read confirms 14 published, 0 drafts, latest v0.8.0. The highest-volume existing descriptions remain v0.5.0 (52,109 chars; 98 top-level bullets), v0.5.2 (17,120; 26), and v0.5.1 (16,406; 35). The candidate-focused note quality guard from PR #565 is merged, but its defined scope leaves historical note bodies unchanged.

    Prioritize v0.5.0's version-specific editor under #536, then the existing #537/#538 batches; preserve real MCP tool permissions, write safety, client transports, required upgrade steps and verified human credits. After approved CHANGELOG edits, apply only separately authorized body-only GitHub Release changes with backups, last-moment source/remote drift checks and readbacks. Complete the 14/14 factual/editorial + public equality audit in #540. Keep skipped-version provenance (#541) separate; do not invent v0.2.0 or v0.5.3 releases.

    This is an evidence update, not a Release mutation or a claim the historical publication work is complete.

  3. yeongseon commented on Nov 21, 2026

    @yeongseon
    ContributorAuthor

    Published-note editorial gate — 2026-11-21 UTC (read-only)

    Fresh GET of all 14 public MCP Server Release bodies shows 9 exact canonical/main matches and 5 substantive divergences (v0.1.0, v0.2.1, v0.3.0, v0.3.1, v0.4.0). This is a separate check from editorial quality, tagged feature truth and contributor provenance.

    An independent structural pass of the public bodies identifies a reader-summary gap beyond those five mismatches: 12 versions (v0.3.0–v0.8.0, excluding nonexistent numbers) begin with a ### category and no user-impact introduction before it. The public v0.2.1 body lacks standard ### categories and largely points readers back to trackers; v0.1.0 has an introductory sentence but legacy nonstandard ### New tools. More importantly, v0.5.0 is 52,109 characters, and v0.5.1/v0.5.2/v0.5.4 also exceed 16 KB: these need content-preserving human editorial review, not a blind length cap.

    Current-head CI is green for source #568, audit #569, and draft manual body-only publisher #581, but no independent PR approval or public-body publication was observed. These proposals must not close the version-specific issues or #540. Keep the disjoint work packages (#531/#536/#537/#538/#530), preserve all permission/security/upgrade limits, and verify actual tagged PR authors before crediting them. The checker/format findings do not prove any statement wrong in released code.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions