commentmd turns a Markdown file into a browser-based review session that returns structured human feedback the agent can act on.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install commentmd
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install commentmd using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is commentmd?
commentmd is an Openclaw Skills workflow for human-in-the-loop Markdown review. It launches a local browser session for design docs, tech specs, READMEs, and PR descriptions so a reviewer can highlight exact text, leave targeted comments, and submit structured JSON back to the agent.
This makes Openclaw Skills especially valuable when an AI agent has already drafted a document but needs precise human guidance before moving forward. Instead of unstructured chat feedback, commentmd produces anchored comments with stable text context, giving the agent a reliable revision signal and a clear audit trail of reviewer intent.
commentmd Use Cases
- The agent has generated a Markdown artifact and needs targeted human review before iterating.
- A user explicitly runs
/commentmd <md_path>on a specific Markdown file. - You want text-anchored feedback for design docs, tech specs, README files, or PR descriptions.
- You need a review flow that works in a local browser, remote machine, or headless environment.
- You want the agent to revise the document and summarize how each comment was addressed.
- You need Openclaw Skills to convert qualitative feedback into structured, machine-readable JSON.
How commentmd Works
- Resolve the provided Markdown path to an absolute file path.
- Start the local review server with
python3 ~/.agents/skills/commentmd/scripts/serve.py "$ABS". - The server binds to
127.0.0.1:3118or the next available port up to3128, then opens the browser for highlight-and-comment review. - The reviewer selects text, adds comments, and clicks Finish to submit the review session.
- The tool writes a structured comments JSON file and exits.
- The agent reads the output file, checks whether the source Markdown changed during review, and then revises the original document.
- The agent replies with a summary of each comment and how it was handled, keeping the Openclaw Skills feedback loop closed.
commentmd Setup
No separate package installation is required beyond Python 3 and a browser; the skill is executed directly from the local skills directory.
- Ensure the skill is available at
~/.agents/skills/commentmd/and thatscripts/serve.pyis executable. - For an interactive local review, run:
python3 ~/.agents/skills/commentmd/scripts/serve.py "/absolute/path/to/document.md"
- For remote or headless environments, generate a static review page:
python3 ~/.agents/skills/commentmd/scripts/serve.py "/absolute/path/to/document.md" --static /tmp/review.html
- Open the generated HTML on the reviewer’s machine, complete the review, and download the JSON file produced by Finish.
- If using the
/commentmd <md_path>command, the agent should resolve the path automatically before launching the review server.
commentmd Data Schema & Taxonomy
commentmd uses a two-file review flow: the source Markdown document and a generated comments JSON file.
| Artifact | Purpose | Location / Naming |
|---|---|---|
| Source Markdown | The document being reviewed | User-supplied <md_path> |
| Review output JSON | Structured review payload returned to the agent | Default: <md_name_without_ext>.comments.json next to the source file |
| Static review HTML | Offline review surface for remote environments | User-defined path such as /tmp/review.html |
Output JSON taxonomy
| Field | Type | Meaning |
|---|---|---|
schema_version |
integer | Fixed schema version for compatibility |
md_file |
string | Absolute path to the reviewed Markdown file |
md_sha256 |
string | SHA-256 hash of the Markdown content captured at server start |
md_changed_during_review |
boolean | Indicates whether the source file was modified externally during review |
created_at |
string | ISO 8601 UTC timestamp for the review submission |
comment_count |
integer | Number of comments submitted; 0 means no objections |
comments |
array | Ordered list of review comments |
Comment object fields
| Field | Type | Meaning |
|---|---|---|
id |
string | Stable comment identifier such as c1 |
quote |
string | The exact excerpt selected by the reviewer |
prefix |
string | 32-character anchor window before the quote |
suffix |
string | 32-character anchor window after the quote |
comment |
string | Reviewer feedback for that excerpt |
created_at |
string | ISO 8601 UTC timestamp for the comment |
The prefix and suffix anchors let the agent map feedback back to the correct part of the document even when surrounding content shifts slightly, which is a core reliability feature of Openclaw Skills.
commentmd Advanced Features
- Local browser launch with automatic HTTP serving on
127.0.0.1for fast review loops. - Static/offline review mode via
--static /tmp/review.htmlfor remote or headless workflows. - Structured JSON output that the agent can parse deterministically without freeform text handling.
- Anchor-based comment capture using
quote,prefix, andsuffixfor precise document targeting. - Source-change detection through
md_changed_during_reviewand content hashing. - Explicit no-change path with
comment_count: 0to signal approval or no objections. - Seamless handoff from Markdown generation to human review to iterative revision in Openclaw Skills.
SKILL.md
Loading
Related Openclaw Skills
commentmd
commentmd lets an AI agent open a Markdown review session in the browser, collect structured human comments, and revise the document with precision.
Break Watch
A Python-powered A-share stock market scanner designed to identify real-time breakout patterns combining volume surges, upward price movements, and bullish candles.
Browser Search
A multi-layered web search and browsing engine that scales from lightweight SearXNG queries to advanced CloakBrowser anti-bot evasion.
Drinks Sommelier
An intelligent, image-enabled AI sommelier that evaluates beer and wine selections against a dynamically updated profile of your personal taste preferences.
commentmd
commentmd turns any Markdown file into a browser-based review session that returns structured human comments the agent can act on.
Publisher Skill
An AI-agent tool designed to publish local HTML frontend projects to PushWebly and package them into installable Android APKs.
Featured*
