Agent context

Updated: July 2026

Agent context is the structured information an AI agent needs to understand a situation and act, such as what happened on screen, when, and where.

An agent-readable recording is one source of agent context: it turns a screen recording into a markdown twin the agent can read. Formats differ in how much context they carry. Video carries almost none an agent can use, a screenshot freezes one instant, logs carry events without the visual flow. The AREC (Agent Recording) document carries the summary, key moments, click targets, and transcript together.

For example, to fix the Export CSV bug the agent needs the failing control, the console error, and the order of events. The markdown twin delivers all three as context in one fetch. The MCP one-call bundle for this is the get_agent_context tool.

Public share links supply context with the Clipy skill alone, no key required.

Related terms: markdown twin, agent legibility, click target.

Learn more: agent context formats compared. See the AREC spec.