What is agent-readable feedback? Definition for AI workflows
Agent-readable means a format an LLM-powered agent can parse and act on without preprocessing — plain text or markdown, with images inline, no JavaScript-rendered UI required.
Agent-readable means a format an LLM-powered agent can parse and act on without preprocessing — plain text or markdown, with images inline, no JavaScript-rendered UI required.
The format requirement comes from how agents consume input. They read text tokens; that is the entire interface. A bug report stored as an interactive Jira ticket is partly invisible because the comments are loaded asynchronously. The same bug report exported as markdown is fully visible because the text is in the file.
What makes a good agent-readable artifact
A single document, top-to-bottom readable in order. Headings that name what each section is about. Screenshots inline rather than linked off-domain. Source URLs on their own lines so the agent can extract and act on them. The pattern works for humans too — the same discipline that makes feedback easy for an agent to act on makes it easy for a teammate three weeks later.
CobaltCapture's markdown export is built around this. A captured review is published as a hosted page and is also one fetch away from a markdown file an agent can ingest without conversion.
Frequently asked questions
What's not agent-readable?
Anything that requires interaction to reveal — collapsed comment threads, JavaScript-rendered overlays, video without a transcript, screenshots living only in a chat backscroll. The agent reads what is in front of it; hidden content stays hidden.
Is HTML agent-readable?
Yes, but noisily. Most agents can parse HTML, but a clean markdown document is shorter, easier to attend to, and produces better results per token.
Capture your first review.
About a minute from open tab to a shareable URL your agent can ingest.
Start capturing