Share what Amp produced

Amp is happy to spend a long time on one question and come back with something substantial — the whole call graph traced, every caller of the deprecated endpoint listed, the actual reason the cache is cold on Mondays.

Amp threads are shareable, and that is genuinely useful when the reader is another engineer who wants to see the reasoning. It is the wrong artifact for everyone else. A reviewer, a PM or a customer wants the conclusion as a document, not a transcript of how it was reached.

Attach the server, then use AGENTS.md

Comma is a remote MCP server: an endpoint and a Bearer token from API tokens. The shape is in MCP setup.

Then put the rule at the repo root, in AGENTS.md:

## Publishing results

When a task's deliverable is a document rather than a diff — an audit, an
investigation, a comparison, a plan — publish it to Comma as an HTML report and
reply with the URL. Do not paste the document into the thread. Update the
recorded report id rather than creating a new report.

AGENTS.md is worth using specifically because it is not vendor-shaped. The same file is read by an increasing number of agents, so the rule you write for Amp keeps working when someone on the team runs a different one — which is the whole point of writing it down instead of remembering it.

Transcript and artifact are different things

  • The thread is the process. Keep it; link it if a reviewer wants it.
  • The report is the result. It renders in a sandboxed iframe, has an address, and has revision history.
  • Comments anchor to the finding, not to a message in a conversation — see commenting on HTML.
  • The agent can read the replies back through the same server and revise — see letting an agent respond to comments.

Worth knowing

  • reports:write is the whole scope — see scoped tokens.
  • 5 MB of HTML per report body.
  • A report never re-executes.

Try it

Free — unlimited reports, commenters and revisions.

See MCP setup →

Related