Make the cost of reading the code a goal, and answer to a reader panel each PR

AGENTS.md

@@ -11,6 +11,10 @@
 - Gitea queues Actions runs that `/repos/{owner}/{repo}/actions/tasks` does not
   list, so an empty task list says nothing about whether CI ran; read
   `/repos/{owner}/{repo}/commits/<sha>/status`.
+- Every pull request scores with the `comprehension-panel` skill's four-seat run at
+  depth 1 and answers it in the same PR, per README goal 13: at or above 7.0, address
+  the findings in one run; below it, fix what the panel names within this PR's reach
+  and re-score, looping while the score climbs. No merge lowers the last score.
 - Hard tabs. No comment by default; delete a restatement, a rationale, history, or a file preamble.
 - One spelling per result: reject the other at `New`, and let the error name the spelling to use.
 - A standing choice a reader would relitigate goes under Decisions in the README, not in a comment.

README.md

@@ -839,6 +839,12 @@ the Development section below, and who ships a register the four above then draw
     nothing, each on its own path. A fixture that looks nothing like production
     tests nothing; the caller who needs a value that can touch nothing asks for it
     by name.
+13. **A cold reader lands in the right place** — the code is judged on what it costs
+    to read, not only on what it does: reaching the unit behind a symptom without
+    asking a person, changing one piece without holding the rest, a layout whose
+    names tell the truth, and a file that stands with no second document open beside
+    it. A simulated reader panel scores those four, the score is a ratchet no release
+    lowers, and [`AGENTS.md`](AGENTS.md) says how each pull request answers to it.
 
 ## Decisions