Skip to content

Releases: wei18/Upkeep

Upkeep v2 — now a Claude Code plugin

Choose a tag to compare

@wei18 wei18 released this 11 Jun 17:33
0d23979

Upkeep is now installable as a Claude Code plugin (/plugin marketplace add wei18/upkeep + /plugin install upkeep@upkeep) or via npx skills add wei18/upkeep --skill upkeep-audit. The GHA reusable workflow is unchanged — @v1 callers keep working but v1 is now frozen; switch to @v2 (identical interface). From now on fixes move the v2 tag only.

Upkeep v2.1.1

Choose a tag to compare

@wei18 wei18 released this 06 Jul 13:32
a4e9a7d

Patch release for GitHub Marketplace publishing.

  • action.yml description shortened under the Marketplace 125-character limit (#22).
  • Plugin version bumped to 2.1.1.

v2.1.0 highlights (composite action.yml for Marketplace, README usage section, docs updates) are in the v2.1.0 notes.

Full Changelog: v2.1.0...v2.1.1

🤖 Generated with Claude Code

Upkeep v2.1.0 — GitHub Marketplace action

Choose a tag to compare

@wei18 wei18 released this 06 Jul 13:22
b7d90da

What's new

  • Composite action at repo root (action.yml) — run the full Upkeep audit as a single - uses: wei18/upkeep@v2 step (#20). This is a sequential alternative built for GitHub Marketplace listing; the reusable workflow remains the recommended integration path when reviewer parallelism matters (see docs/en/why-reusable-workflow.md).
  • README: new "Or as a Marketplace action" usage section.
  • Docs: the composite-action trade-off is now recorded in all five language versions of why-reusable-workflow.md.
  • Plugin version bumped to 2.1.0 (#21).

Full Changelog: 64745c7...v2.1.0

🤖 Generated with Claude Code

Upkeep v1

Choose a tag to compare

@wei18 wei18 released this 05 Jun 18:21
89fe8e0

Upkeep — your AI writes fast, Upkeep keeps it honest

Upkeep is a reusable GitHub Action that audits a repository for drift — stale docs, orphaned files, convention violations, spec/flow mismatches — by dispatching a team of specialized AI reviewers (powered by Claude). It treats your repo's own conventions as the rubric, reports divergence with evidence + severity, and never auto-edits.

What it checks

Seven reviewers (i18n opt-in):

  • docs_staleness — outdated docs/READMEs, incl. multilingual-sync drift
  • code_hygiene — dead code, leftovers, smells
  • spec_flow — spec ↔ implementation/flow divergence
  • visual_icon — stale or duplicate visuals & icons
  • duplicate_orphan — duplicate & unreferenced files
  • convention — your repo's own CLAUDE.md / skills / workflows as the rubric
  • i18n — localization completeness (opt-in)

Usage

jobs:
  audit:
    uses: wei18/upkeep/.github/workflows/audit.yml@v1
    permissions:
      contents: read
      issues: write
      id-token: write
    secrets:
      claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}

Set the repo secret with claude setup-token (uses your Claude Pro/Max plan — no separate API billing).

Outputs

  • Self-contained HTML report artifact
  • One upserted GitHub tracking issue (label audit)

Options

model · max_turns · issue_label · rubric_lang (reviewer language: en | zh-TW)

Docs

README + design docs in English · 繁體中文 · 简体中文 · 日本語 · 한국어