Use case
Agent-ready feedback with Patchrooms
Feedback formatted for a coding agent, not just a human. Every report carries the artifact metadata (goal, constraints, tool) plus the element selector and a screenshot, so agents like Claude Code, Cursor, and Codex can act without asking questions.
Try the live demoThe problem
Coding agents are only as good as the context you give them. A screenshot and "fix this" forces the agent to guess the target and the intent, which leads to wrong fixes and wasted generations.
The workflow
-
Capture with context
In artifact-review mode, Patchrooms attaches goal, constraints, and tool to every comment alongside the selector and screenshot.
-
Export as agent-ready Markdown
The report is structured so an agent can read intent, target, and constraints at a glance.
-
Agent fixes precisely
The agent edits the right element to satisfy the stated goal and constraints, with no clarifying round trip.
Example
A report on a checkout page carries the goal ("reduce friction"), a constraint ("keep the layout single-column"), and the selector for the coupon field. Claude Code reads it and removes the field without breaking the layout.
Agent-ready export
Reports export as Markdown that you paste into your agent, or agents read them directly over the Patchrooms MCP server (list_reports / get_report, authed with pr_sk_* keys), available now. Status writes (set_status) are available now; extended MCP tools are planned.
A report exports as Markdown like this:
## Report: Checkout review
**Tool:** claude-code **Goal:** reduce checkout friction
**Constraint:** keep the layout single-column
- **Element:** `input#coupon-code`
- **Comment:** Remove the coupon field; almost no one uses it.
- **Screenshot:** blob/scr_5da7.png FAQ
- What makes feedback "agent-ready"?
- It carries structured context an agent can act on: the element selector, a screenshot, and the artifact goal and constraints, not just a free-text description.
- Can agents read Patchrooms reports automatically?
- The Patchrooms MCP server for direct reads (list_reports / get_report) is available now. You can also export reports as Markdown and paste them into your agent. Status writes (set_status) are available now; extended MCP tools are planned.
- Which agents can consume it?
- Any agent that accepts text context: Claude Code, Cursor, Codex, Windsurf, OpenCode, and Cline all read the exported Markdown directly.