One install for the author

So you are staring at another CLI and wondering if it is worth the tax. The agent is mid-plan. You want a teammate on it. The reviewer path, if this product is doing its job, is a browser and a link. The author path is install. That asymmetry is not a packaging accident. It is the product.

If the reviewer also has to install something, you lost before the first comment. You traded “open the link” for a fleet of setups, and the loop never starts.

What people do instead of paying that tax

Solo in the same IDE or agent session is the zero-ceremony option. It is also zero second human. No durable thread outside the session. Fine when you truly do not need outside eyes. Less fine the moment you do.

Screenshare skips a new install for either side. You pay a calendar instead. You get high bandwidth once, verbal nuance, shared cursor. You do not get durable line-anchored threads the writing agent can act on after you hang up. The conversation evaporates with the call.

Slack paste is zero install and wins on interrupt. The paste freezes. The agent does not get a structured inbox on the live file. Someone still copies notes into the prompt and hopes the right paragraph survives the paste.

Early PR feels familiar because GitHub review UI is muscle memory. Both of you are already in git. That is the wrong timeline for a plan the agent is still allowed to rewrite. Before-git review and after-commit review are different jobs. Do not open a PR just so someone can comment on text that is still allowed to thrash.

Those substitutes win when nobody will install anything, or when a frozen chat blob is enough. They lose when you want a live hosted watch, Docs-style comments, and agent connect into the session that owns plan.md. The install question is really a loop question: will you pay once on the author side so the reviewer path stays empty?

The install tax belongs on one side

Author installs. Reviewer does not.

That is the product claim, not a slogan. The person who owns the agent session pays the setup cost. The person leaving comments opens a URL. No IDE. No repo clone. No agent account on the reviewer side.

I will say this as advocacy, not as a setup-funnel study: putting install on the reviewer kills the loop before it starts. You are asking every teammate to adopt tooling for a job that should be “open the link.” The author already lives with the agent, the file, and the machine. One install there is the right tax. Spreading that tax across every reviewer is how you end up back in Slack paste and early PRs.

What the author is actually installing for

CLI gets you onto the loop. After install you can run a hosted watch so saves update a live page. Local preview (gander file.md --watch) is not the same as hosted watch (gander watch file.md). Reviewers need the hosted link. Mixing those up is how people think they “shared” and still left the teammate staring at a local-only preview.

Skill and MCP are the wiring behind agent connect, so a comment that starts with @agent lands in the agent session. MCP is preferred for agent connect. gander comments is the fallback. Without that wiring you can still share a live page and collect human threads, which is useful. You do not have agent connect. Human-only threads stay human-only.

Homebrew installs the binary. Skill and MCP are separate steps (gander skill, gander mcp install). Curl install path and Homebrew both exist. Platforms are macOS and Linux, Windows via WSL. Exact commands live on the install docs, not at the end of this essay, because this is not a try-it closer disguised as a blog post.

After that setup, the product is the loop: watch, comment, @agent, same link updates, then git. Before git, not instead of git. Full loop is shipped. Local file stays source of truth. Hosted page is live view plus threads. You are not buying a prettier markdown preview. Preview is a piece. The product is live review for a plan the agent is still writing.

Not the browser-only post, and not the comment-prefix post

An earlier cut said the reviewer should not need your IDE. That was capability.

This cut is entry: how the author gets into the loop so that capability is real. Skill and MCP are not a second product story. They are how comments reach the session that is rewriting the file.

Do not flatten this into a post about the @agent prefix either. That one is about the comment syntax and the inbox. This post is who pays install so that inbox exists.

Same beachhead. Different spine. Reviewer still only needs a browser and a sign-in. Author still pays once so the loop can run.

What this is not

Not an install tutorial. Not a try-it closer with curl fences pasted under a soft CTA. Not skills.sh listed. Not invented install counts or stars. Not “easy setup thousands use.” Not skill frontmatter as marketing copy until product rewrites it.

Not share-markdown as the lead. Not for PMs. Not replace GitHub or Notion. Not a preview toy with a download button at the bottom.

If you need the one-liners, they are on gander.md/docs/install. This essay’s job is the asymmetry, not the cheatsheet.

Who pays

The next time “another tool to install” feels like friction, ask who would have to install it.

If the answer is every reviewer, the loop is already dead. You traded a browser link for a rollout project. If the answer is the author once, so a teammate can review from a browser and agent connect can reach the session, that is the tax the product asks for. It is a small tax compared with asking every commenter to join your agent stack.

One install for the author. Zero for the reviewer. CLI on the loop. Skill and MCP for the agent. Then watch, comment, @agent, then git.

The install belongs on the side that owns the agent.