Comment Mode: Granular Feedback Without Losing the Writer's Voice

Where this fits

You've just written a 2,000-word product announcement. You ask your agent to "make it sharper" and get back a rewrite that sounds like someone else. That's the pain comment-mode solves: instead of a full rewrite, it produces highlighted HTML with click-to-reveal comments. You keep your prose, but you get concrete, line-level feedback you can accept or dismiss one piece at a time—exactly like a human editor scribbling in the margins.

Why agents benefit

• Preserves user authorship: Agents act as critics, not ghostwriters. The draft stays yours; feedback appears as annotations, so the user retains full control over changes.
• Multi-lens analysis: The skill can simulate a specific person's reaction ("as brian would react") or focus on a single dimension ("word choice only"), letting agents tailor feedback depth and style without extra prompts.
• Structured output for review: Generates a self-contained HTML file with click-to-reveal overlays—no need for viewers to parse a chat transcript or a list of comments. It's a shareable artifact.
• Fits into existing agent pipelines: The skill uses only Bash(open *) and Write, so it's lightweight and easy to drop into any workflow—no external services or model-dependent behavior (model invocation is disabled).
• Enables incremental iteration: Since feedback is granular and non‑destructive, users can revise in waves: address one highlighted paragraph, then ask for a second pass on remaining issues.

Practical scenarios

Editorial pass on a blog draft

You paste a draft and say "comment on this like a sharp editor." The agent returns an HTML file with notes on flow, redundancy, and factual gaps. You click through each comment, revise what you agree with, and ignore the rest—your voice stays intact.

Simulating a stakeholder's response

Before sending a proposal to your boss Brian, you ask the agent to "leave comments as Brian would react." The skill generates annotations like "this budget line needs a backup" or "tighten the timeline here," giving you foresight into potential pushback without a real meeting.

Targeted proofreading (“word choice only”)

You're happy with the structure but suspect lazy adjectives. A single phrase—"focus on word choice"—produces a comment file that flags only weak or vague terms, letting you polish language without re‑evaluating the whole argument.

Add it to your agent workflow

Install the skill with:

npx skills add shipshitdev/skills --skill comment-mode

Then, in your agent, reference the draft and desired lens. For example:

Comment on /path/to/draft.md as if you were a skeptical investor. Save the output to _private/views/investor-comments-temp.html.

The agent generates the HTML, opens it in your browser, and you're ready to review. For a full walkthrough, see the skill details on OpenAgentSkill.

Compare before adopting

Before making comment-mode a staple, compare it against alternatives by checking:

• Quality signals: Are the comments genuinely useful, or do they read like generic filler? Test with your own drafts.
• Maintenance freshness: Look at the repository's last commit and issue activity. A skill that isn't maintained may break with agent updates.
• Alternatives: Consider direct rewrites, checklist-based review, or other feedback skills—decide which matches your workflow best.
• Workflow fit: Does the HTML output integrate with your usual review tools? If you often work in Markdown or Google Docs, the click-to-reveal format may or may not align.

Why it is worth tracking

With 33 stars and a clear, focused implementation, comment-mode is a promising pattern for non‑destructive feedback. Its design—granular, lens‑driven, and output‑as‑artifact—addresses a real pain point that many rewrite‑happy agents ignore. Track it for how the community adopts it, whether the skill evolves to support new lenses or export formats, and how it fares in real‑world feedback loops. Even if it doesn't become a daily‑driver, it's worth watching as a reference for building your own feedback mechanisms.