docs(agents): a PR with no intent line shows the prompt that produced it

AGENTS.md

@@ -13,9 +13,12 @@ Code is checked; a discussion is trusted. Mark which is which —
    footer. Verbose evidence goes in `<details>`. Applies to PR bodies, issues
    and comments posted from a session.
 2. **The intent line is the human's**, unless they ask you to write the whole
-   thing — then write it. Usually they add a few sentences at the top; quote the
-   ask verbatim rather than paraphrasing it into first-person motivation, and if
-   the slot is empty, leave it empty and say so.
+   thing — then write it. Usually they add a few sentences at the top. Never
+   paraphrase their ask into a first-person "I want this because…".
+   **Where they wrote no intent, quote the prompt in its place** — verbatim,
+   labelled as the prompt, and above the note; several prompts, several quotes,
+   in the order they were given. An empty slot tells a reader nothing, while the
+   prompt tells them exactly what was asked for, in the human's own words.
 3. **Do not hold the conversation.** Post marked information — a log, a
    benchmark, a diff. Do not reply, concede, agree or decide as someone else.
 
@@ -35,6 +38,15 @@ What was implemented, and how it was verified.
 </details>
 ```
 
+Where that first line was never written, the prompt stands in for it:
+
+```markdown
+> **Prompt:** Do a PR applying the instructions from AGENTS.md to the entire repo.
+
+> [!NOTE]
+> The following content was generated by AI.
+```
+
 In the tree nothing is marked: `Co-Authored-By: Claude …` is the record.
 
 ## Code