md-review

Name: md-review
Author: alirezarezvani/claude-skills

$npx mdskill add alirezarezvani/claude-skills/md-review

Converts markdown PR writeups to a single-file HTML code review

Solves the problem of converting markdown PR writeups with diff blocks and severity tags into a structured HTML review.
Depends on `diff_parser.py`, `annotation_extractor.py`, and `review_html_renderer.py` for processing and rendering.
Decides based on input classification by `markdown-html-orchestrator` or direct invocation, ensuring proper formatting and accessibility.
Delivers the rendered HTML code review to users via a jump-nav layout with severity-tagged annotation cards.

SKILL.md

.github/skills/md-reviewView on GitHub ↗

---
name: md-review
description: Converts a markdown PR writeup or code review (one with ```diff fenced blocks and severity-tagged > [!BLOCKER]/[!MAJOR]/[!MINOR]/[!NIT] callouts) into a single-file 2-column HTML review — unified-diff on the left, severity-tagged annotation cards on the right, top jump-nav listing every finding, mandatory named reviewer footer. Triggers when the markdown-html-orchestrator classifies an input as REVIEW, or when invoked directly via /cs:md-review. Refuses without explicit --reviewer (a code review must name a human), refuses if no diff hunks present (route to md-document instead), and refuses to encode severity in color only (every badge ships color + icon + aria-label per WCAG 1.4.1). Use after orchestrator routing.
version: 2.10.2
author: Alireza Rezvani
license: MIT
tags: [markdown, html, code-review, diff, severity, annotations, single-file, design-system, wcag-1.4.1]
compatible_tools: [claude-code, codex-cli, cursor, antigravity, opencode, gemini-cli]
---

# md-review — Code-review markdown → 2-column HTML

The code-review converter from Tier 2 of Shihipar's essay ("Code Review and PR Writeups"). Takes a markdown PR writeup with diff blocks + severity callouts and produces a single-file HTML review with a jump-nav, 2-column diff + annotation layout, and a named reviewer footer.

Three stdlib tools pipeline together:

```
diff_parser.py  →  annotation_extractor.py  →  review_html_renderer.py
   (md → diff hunks)   (md → severity-tagged    (hunks + annotations
                        annotations attached     + tokens → 2-col HTML)
                        to nearest hunk)
```

## When to invoke

| Symptom | Action |
|---|---|
| `markdown-html-orchestrator` routes input as REVIEW | Invoke this skill |
| User runs `/cs:md-review <path>.md` directly | Invoke this skill |
| Input contains ` ```diff ` fenced blocks + `> [!MAJOR]`/`> [!BLOCKER]`/etc. callouts | Invoke this skill |
| Input is a long-form spec / report (no diff blocks) | Route to `md-document` instead |
| Input is a slide deck | Route to `md-slides` instead |
| Input < 100 lines | Refuse (Shihipar threshold) |
| Design-system not onboarded | Refuse; surface `/cs:design-system` |

## Pipeline

```bash
# 1. Parse markdown → diff hunks JSON
python3 markdown-html/skills/md-review/scripts/diff_parser.py \
    --input <path>.md --output hunks.json

# 2. Extract severity-tagged annotations, attach to nearest preceding hunk
python3 markdown-html/skills/md-review/scripts/annotation_extractor.py \
    --input <path>.md --diff-blocks hunks.json --output annotations.json

# 3. Render 2-col HTML (--reviewer is mandatory — refuses without)
python3 markdown-html/skills/md-review/scripts/review_html_renderer.py \
    --diff-blocks hunks.json --annotations annotations.json \
    --reviewer "Jane Doe" --title "PR #123: Add retry logic" \
    --output review.html
```

## What gets rendered

- **Top jump-nav** — every annotation with severity badge + 80-char preview + jump link; severity counts in the heading ("3 BLOCKER · 2 MAJOR · 1 NIT")
- **2-column hunk rows** — unified diff on the left (per-line old/new line numbers, +/− marks, addition/deletion background tint from design-system tokens), annotation cards on the right (color + icon + aria-label per WCAG 1.4.1)
- **Approval bar** — if `LGTM` markers are present and no severity annotations, a success-tinted "LGTM — no findings flagged" bar
- **General comments** — annotations not attached to any hunk render at the bottom in their own section
- **Reviewer footer** — mandatory; refuses to render without `--reviewer`
- **Responsive** — 2-col collapses to stacked on viewports < 900px

## Hard rules

1. **`--reviewer` is mandatory.** A code review must name a human reviewer. Refuses with exit 3 otherwise. Mirrors research-ops's "named owner" discipline.
2. **Refuses if no hunks present.** No `--- a/file` + `@@ ... @@` blocks means this isn't a code review — refuses with exit 4 and recommends `md-document`.
3. **Refuses input < 100 lines.** Markdown wins below the threshold (Shihipar).
4. **Refuses without onboarding.** Same gate as every converter.
5. **Severity is never color-only.** Each badge ships color + icon + `aria-label` + text. WCAG 1.4.1 enforced at the renderer level.
6. **Single-file output.** All CSS inline. Only external is Google Fonts CSS. No Prism in md-review (diff coloring conflicts with syntax highlighting).
7. **Custom severity convention.** `--severity-convention "critical,important,suggestion,nit"` swaps tier names; position 0 is most severe. Default is BLOCKER / MAJOR / MINOR / NIT (Google Code Review Developer Guide).

## Forcing-question library (Matt Pocock grill discipline)

1. **Who is the named reviewer?** Recommended: the user signing off on the review. Canon: research-ops named-owner pattern; *SWE at Google* ch. 9.
2. **Which severity convention applies — default (BLOCKER/MAJOR/MINOR/NIT) or custom?** Recommended: default unless your team has a documented alternative. Canon: Google *Code Review Developer Guide*.
3. **Are annotations anchored to specific hunks, or are some general?** Recommended: anchor everything you can; general goes to the unanchored section. Canon: *SWE at Google* ch. 9 — "Comments must reference a specific line".
4. **What's the PR title for the `<title>` and header?** Recommended: the actual PR / commit title. Canon: docs-as-context-for-readers.
5. **Should `LGTM` markers ship as the approval bar?** Recommended: yes if there are no severity annotations; otherwise the findings take precedence.

## Distinct from

- **`md-document`** — that converter renders prose + tables + code + callouts. This one renders diff hunks + margin annotations.
- **`md-slides`** — that converter splits on `---` boundaries. This one is a single-page artifact.
- **GitHub PR comments** — those are a thread. This is a single-author snapshot artifact.

## Output artifact

`{default_output_dir}/review-{slug}.html` (path resolved by orchestrator's `output_path_resolver.py`; collision suffix `-2`, `-3`, … by default).

## References

- Shihipar — *Claude Code HTML output* (Medium, 2026), Tier 2 use case
- *Software Engineering at Google* (Manshreck & Wright, O'Reilly 2020), ch. 9 "Code Review"
- Google *Code Review Developer Guide* — severity convention source
- WCAG 2.2 §1.4.1 — color-not-sole-signal enforcement
- See `references/` for full citations (diff_rendering_canon, severity_coding, pr_annotation_ux)

More from alirezarezvani/claude-skills

Skill	Description
a11y-audit	Accessibility audit skill for scanning, fixing, and verifying WCAG 2.2 Level A and AA compliance across React, Next.js, Vue, Angular, Svelte, and plain HTML codebases. Use when auditing accessibility, fixing a11y violations, checking color contrast, generating compliance reports, or integrating accessibility checks into CI/CD pipelines.
ab-test-setup	When the user wants to plan, design, or implement an A/B test or experiment. Also use when the user mentions "A/B test," "split test," "experiment," "test this change," "variant copy," "multivariate test," "hypothesis," "conversion experiment," "statistical significance," or "test this." For tracking implementation, see analytics-tracking.
ad-creative	When the user needs to generate, iterate, or scale ad creative for paid advertising. Use when they say 'write ad copy,' 'generate headlines,' 'create ad variations,' 'bulk creative,' 'iterate on ads,' 'ad copy validation,' 'RSA headlines,' 'Meta ad copy,' 'LinkedIn ad,' or 'creative testing.' This is pure creative production — distinct from paid-ads (campaign strategy). Use ad-creative when you need the copy, not the campaign plan.
adversarial-reviewer	Adversarial code review that breaks the self-review monoculture. Use when you want a genuinely critical review of recent changes, before merging a PR, or when you suspect Claude is being too agreeable about code quality. Forces perspective shifts through hostile reviewer personas that catch blind spots the author's mental model shares with the reviewer.
aeo	Answer Engine Optimization (AEO) skill — optimize content to be cited by AI language models (ChatGPT, Perplexity, Claude, Gemini, Mistral) as authoritative sources. Distinct from SEO — AEO optimizes for citation in LLM-generated responses, not search rankings. Use when planning content for AI-first search audiences, auditing existing content for E-E-A-T signals, tracking which pages get cited by which LLMs, or building a citation-friendly content strategy. Triggers — 'AEO audit', 'optimize for ChatGPT', 'get cited by Perplexity', 'LLM citation strategy', 'answer engine optimization', 'content for AI search', 'E-E-A-T audit'. Output is a markdown audit report (default) or JSON for pipeline integration. Stdlib-only Python tools.
agent-decision-receipts	Mint a tamper-evident, post-quantum-signed receipt for a consequential agent action (deploy, delete, pay, grant-access, model decision) so it can be verified later from the certificate alone. Use when an autonomous agent takes a side-effecting action that may need to be proven later, or when satisfying EU AI Act Article 12 record-keeping. Three decisions: whether an action needs a receipt, minting it, verifying it. Signing is delegated to the open-source OpenAgentOntology package. Not after-the-fact log analysis; not a hosted notary; not a legal opinion.
agent-designer	Use when the user asks to design a multi-agent system, pick an orchestration pattern (supervisor/swarm/pipeline), generate tool schemas for agents, or evaluate agent execution logs for cost, latency, and failure bottlenecks. Examples: 'design an agent architecture for research automation', 'generate Anthropic tool schemas from these tool descriptions', 'analyze these agent run logs for bottlenecks'. NOT for Claude Code workflow files (use workflow-builder) or single-agent prompt design (use agent-workflow-designer).
agent-harness	Turn any domain folder of skills into a bounded agentic loop: compile a goal into a verifiable task plan, execute tasks with the domain's own tools, verify every task with machine-run checks, retry with caps, escalate to a human when budgets exhaust, and refuse to close until everything is verified or explicitly waived. Use when you want an agent or subagent to pick up a goal and drive it to a verified close across one of this repo's 18 domains ('run this goal through the engineering harness', 'set up an agentic loop for marketing work', 'make the finance domain self-verifying'). NOT for authoring Claude Code Workflow-tool .js scripts (workflow-builder), N-agent tournaments on one task (agenthub), single-file metric optimization (autoresearch-agent), or discovering published loop recipes (loop-library).
agent-launcher-orchestrator	Use when a user wants to build, launch, grade, or schedule a Claude Managed Agent (CMA) in their own Anthropic account — "build me an agent", "launch this as a managed agent", "run this on a schedule", "grade my agent against a rubric", "set up a nightly worker". Reads the per-session goal (./my-agent/goal.json), routes deterministically to one of five phase sub-skills (interview → stage-launch → grade-iterate → run-without-you → wrap-up) via goal_router.py, and compiles the goal+phase into an execution shape (single-pass workflow / bounded grade→iterate loop / recurring cron deployment loop) via loop_compiler.py. Forks context so heavy intake (build sheets, payloads, eval cases) stays out of the parent thread. All launches are emitted as BYOK curl the user runs with their own key; no tool makes API calls. Inspired by anthropics/launch-your-agent (Apache-2.0). Distinct from engineering/agent-harness (generic domain loop) and engineering/write-a-skill (authors Claude Code skills, not CMAs).
agent-memory	Use when a project's CLAUDE.md has grown past what anyone reads and you want the agent to learn durable facts from its own sessions instead — or when asking why the agent keeps re-learning the same correction, why a remembered rule is wrong, or where a memory line came from. Implements a four-tier store (L0 transcripts / L1 candidates / L2 project context / L3 stable persona) where promotion is earned by recurrence across sessions and days, never by one confident statement, and nothing reaches a committed file without a human adopting it.