Open a browser tab so a human can highlight-and-comment on a Markdown file. Submits a structured JSON of comments the agent can read and respond to. Use when...
浏览器
Commentmd
试用Open a browser tab so a human can highlight-and-comment on a Markdown file. Submits a structured JSON of comments the agent can read and respond to. Use when...
它能做什么
Open a browser tab so a human can highlight-and-comment on a Markdown file. Submits a structured JSON of comments the agent can read and respond to. Use whenever the agent has generated a Markdown document (design doc, tech spec, README, PR description) and needs human review before proceeding.
技能文档
commentmd
When to invoke
- The user explicitly runs
/commentmd. - The agent has generated a Markdown document (design, tech spec, README, PR description, etc.) and needs targeted human feedback before iterating.
Usage
/commentmd
`` is a relative or absolute path to the Markdown file to review.
Agent execution steps
- Resolve `` to an absolute path
$ABS. - Run:
python3 ~/.agents/skills/commentmd/scripts/serve.py "$ABS" - The command starts a local HTTP server (
127.0.0.1:3118, or the next free port up to3128) and opens the browser. The user selects text, adds comments, then clicks 完成评论 (Finish). The command printswroteand exits. - Read `` — by default
.comments.jsonnext to the source file. - Process each comment:
quoteis the original excerpt;prefix/suffixare 32-char anchor windows around it;commentis the human's note. Ifmd_changed_during_reviewistrue, warn the user that the source file was modified externally between server start and submit. - Revise the original document based on the comments and summarize in the reply how each was addressed.
Headless / remote environments
Append --static /tmp/review.html:
python3 ~/.agents/skills/commentmd/scripts/serve.py "$ABS" --static /tmp/review.html
The user opens the HTML on their own machine; Finish downloads a JSON file. Ask the user for the downloaded file's path, then read it.
Output JSON format
{
"schema_version": 1,
"md_file": "/abs/path/plan.md",
"md_sha256": "abc...",
"md_changed_during_review": false,
"created_at": "2026-07-01T10:00:00Z",
"comment_count": 2,
"comments": [
{
"id": "c1",
"quote": "Store events in MySQL",
"prefix": "In our storage layer, we ",
"suffix": ", replicated across ...",
"comment": "Why not PostgreSQL? JSONB support is better.",
"created_at": "2026-07-01T10:00:12Z"
}
]
}
comment_count: 0 means the user has no objections.
相关技能
Open a browser tab so a human can highlight-and-comment on a Markdown file. Submits a structured JSON of comments the agent can read and respond to. Use when...
Turn Markdown files or folders into a polished, human-readable, self-contained HTML page by having the agent understand the document, choose an appropriate i...
Fetch any public web page and get back clean, LLM-ready markdown (polite, robots-respecting)
Retrieve only relevant sections from Markdown memory, project notes, decision logs, and heading-structured knowledge bases with the mdselect CLI. Use when Co...
Aggressively produce self-contained HTML artifacts instead of Markdown for substantial agent outputs. Use when the user asks for a doc, plan, spec, report, r...