| name | pr-scope |
| description | Use when the user wants a PR scope assessment surfaced directly — "is this PR too big", "how should I split this", "chunk this into PRs", "scope check", or before opening a large PR. Runs the pr-scope-analyzer agent and relays its full output verbatim. |
Run the pr-scope-analyzer agent and surface its complete output to the user.
Why this skill exists
When a subagent finishes, its final message is returned to the top-level model as a tool result — the user never sees it directly. Left alone, the top-level model paraphrases the agent's carefully-formatted impact table, verdict, and dependency graph into a lossy summary. This skill exists to stop that: the agent's formatting is the deliverable, and it must reach the user intact.
Steps
-
Dispatch the agent. One Agent call, subagent_type: pr-scope-analyzer, run_in_background: false (you need the result in this turn). In the prompt, tell it what to assess: the current branch's changes by default, or a specific ref/directory if the user named one. The agent computes its own diff and carries its own instructions — keep the prompt to what to look at.
-
Relay verbatim. Your entire response is the agent's final message, reproduced exactly — the impact table, the verdict line, and the split recommendation with its dependency graph, character for character, in the layout the agent produced. Preserve the code fences and alignment. Add nothing before it and change nothing inside it.
After the output
You may add commentary only below the reproduced output, and only if the user asked a follow-up question the agent's report doesn't already answer. If they just asked for a scope check, the reproduced output is the whole response.
