livlign skills

A Claude Code plugin marketplace for open-source maintainers — README heroes, README audits, PR descriptions in your repo's own voice, and trustworthy .NET test-coverage backfills.

/plugin marketplace add livlign/claude-skills

Already added the marketplace? Run /plugin marketplace update livlign to pick up newly-published plugins (e.g. dotnet-coverage-kit) and version bumps.

Plugins

repo-visuals

Scans the target repo, recommends a format that fits its identity, proposes scenarios, designs a bespoke HTML stage, previews it in your browser, and exports a retina-quality artifact — then optionally opens the upstream PR.

Featured in rohitg00/awesome-claude-code-toolkit — the curated list of Claude Code plugins, agents, and MCP servers.

What a run looks like:

> /repo-visuals
  pick mode → semi-auto · scan repo → 30 tools detected
  recommend → animated GIF, 1200×675 · brief locked
  build HTML → preview in browser → ship
  export → hero.gif (2.4 MB) · open PR #42

Why it's different

• Discovery dialog, not templates. Every hero is bespoke. A structured scan-then-design conversation runs before any HTML is written.
• Production-grade by default. Retina capture, drift-proof copy (no stale star counts or version numbers), preview-vs-export parity, scope-match guardrails — encoded from real maintainer pushback.
• Three modes. Manual, Semi-auto (recommended), Auto.
• Optional PR. Local-only or full upstream PR with relative-path embed, alt-text generation, and fork-and-PR for repos you don't own.

Recent heroes shipped

Thumbnails — click through to see the live hero in each repo.

<table> <tr> <td width="50%" valign="top"> _{<a href="https://github.com/htmlhint/HTMLHint"><b>htmlhint/HTMLHint</b></a>} <a href="https://github.com/htmlhint/HTMLHint"><img src="./assets/showcase/htmlhint.png" alt="HTMLHint hero"></a> </td> <td width="50%" valign="top"> _{<a href="https://github.com/gui-cs/Terminal.Gui"><b>gui-cs/Terminal.Gui</b></a>} <a href="https://github.com/gui-cs/Terminal.Gui"><img src="./assets/showcase/terminal-gui.png" alt="Terminal.Gui hero"></a> </td> </tr> <tr> <td width="50%" valign="top"> _{<a href="https://github.com/kunwar-shah/claudex"><b>kunwar-shah/claudex</b></a>} <a href="https://github.com/kunwar-shah/claudex"><img src="./assets/showcase/claudex.png" alt="claudex hero"></a> </td> <td width="50%" valign="top"> _{<a href="https://github.com/rullerzhou-afk/clawd-on-desk"><b>rullerzhou-afk/clawd-on-desk</b></a>} <a href="https://github.com/rullerzhou-afk/clawd-on-desk"><img src="./assets/showcase/clawd-on-desk.png" alt="clawd-on-desk hero"></a> </td> </tr> <tr> <td width="50%" valign="top"> _{<a href="https://github.com/affaan-m/everything-claude-code"><b>affaan-m/everything-claude-code</b></a>} <a href="https://github.com/affaan-m/everything-claude-code"><img src="./assets/showcase/everything-claude-code.png" alt="everything-claude-code hero"></a> </td> <td width="50%" valign="top"> _{<a href="https://github.com/emtyty/ast-graph"><b>emtyty/ast-graph</b></a>} <a href="https://github.com/emtyty/ast-graph"><img src="./assets/showcase/ast-graph.png" alt="ast-graph hero"></a> </td> </tr> </table>

Full showcase →

Skill source →

repo-doctor

Audits a README the way a senior maintainer would — does the "what is this" sentence land, is install-to-first-success short enough, does the audience-jargon match, are there drift signals (stale versions, badge sprawl, dead links). Read-only diagnostic by default; opens a PR only on explicit ask.

What a run looks like:

> /readme-doctor
  scan repo → classify category (CLI tool · library · framework · …)
  audit → score 9 criteria with evidence
  output → P0/P1/P2 punch list, each with a one-line fix
  health score: 69/100

Skill source →

pr-voice

Rewrites a thin or empty PR description the way this repo already writes them. It profiles the merged-PR history first — title convention, body length for the diff size, the headings that recur and in what order, how issues get closed — then drafts from the diff, commits, and linked issues only. No invented motivation, no "all tests pass" nobody checked, and nothing the author wrote gets dropped.

What a run looks like:

> /pr-voice 42
  target → PR #42 · +38/−9 · body: "fix"
  voice  → 31 merged PRs, bots skipped · fix(scope): titles · ~70 words · Why → What
  draft  → before/after · dropped "Testing" (no evidence tests ran)
  apply? → gh pr edit 42 ✓

Plugin source →

dotnet-coverage-kit

Backfills unit tests into legacy .NET services and keeps coverage honest. It classifies every file per method — so a controller or a DbContext-bound god-class still contributes its testable slice instead of being written off wholesale — writes characterization tests for old code and spec tests for new, and measures deterministic C0/C1 via dotnet-coverage. A checked-in manifest explains why every file is tested, excluded, or can't-be-tested (each with a mitigation), and a CI gate (diff coverage + ratchet + scope guard) stops it slipping.

What a run looks like:

> /coverage-init
  detect projects → parallel sweep classifies every file (per-method carve-outs)
  draft manifest → self-critique loop → you confirm only the gray areas
> /generate-tests
  characterize old code · spec new code → 100% of testable branches
  genuinely untestable → manifest entry with a reason + mitigation
> /coverage-report
  deterministic C0/C1 → merge sets the baseline → CI gate on every PR

Plugin source →

Contributing

Easiest entry points: report a real failure as an issue, pick something from the wishlist, or submit a hero to the showcase.

Full guide → CONTRIBUTING.md

Personal project, evolving fast — expect breaking changes pre-1.0.