Enforce TDD in the implement flow

Add a PreToolUse hook that reminds the main agent not to author test
files directly, and change the implement skill to always use
test-driven-development rather than only where possible.

.claude/settings.json

@@ -19,6 +19,17 @@
           }
         ]
       }
+    ],
+    "PreToolUse": [
+      {
+        "matcher": "Write|Edit",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "f=$(jq -r '.tool_input.file_path // empty'); printf '%s' \"$f\" | grep -Eq '(\\.(test|spec)\\.[cm]?[jt]sx?$)|/__tests__/' && jq -n '{hookSpecificOutput:{hookEventName:\"PreToolUse\",additionalContext:\"test-driven-development skill: the MAIN agent must not author test files. Each test is written by a test-writer sub-agent (via the Agent tool) from the public interface alone. If you are running /implement or any TDD flow and have not loaded /test-driven-development, load it now and delegate this test to the sub-agent. If you ARE the test-writer sub-agent, disregard this reminder.\"}}' || true"
+          }
+        ]
+      }
     ]
   },
   "model": "opus"

.claude/skills/implement/SKILL.md

@@ -30,7 +30,7 @@ Always branch off `main`, never off a sibling task branch.
 
 ### 3. Implement
 
-Build the work described in the task's "What to build" section, satisfying its acceptance criteria. Use `/test-driven-development` where possible, at the seams already agreed when the spec or task was written.
+Build the work described in the task's "What to build" section, satisfying its acceptance criteria. ALWAYS use `/test-driven-development`, at the seams already agreed when the spec or task was written.
 
 Run typechecking regularly, single test files regularly, and the full test suite once at the end.