Playbook 07: a role started more than once is an agent definition in the repository it serves; hq's words test reads .claude/

00-META/checks/words_test.go

@@ -52,7 +52,7 @@ const (
 	linesFrom = "2026-10-09"
 )
 
-var wordsAlways = []string{"00-META/", "03-DESIGN/", "AGENTS.md", "README.md"}
+var wordsAlways = []string{"00-META/", "03-DESIGN/", ".claude/", "AGENTS.md", "README.md"}
 
 // daemonLines are the daemons' lines (ADR 0272): each daemon's commands. "alone" is a command that is a line by
 // itself; "with an argument" one that is a line only when written as one ("the mesh-controller module" is prose;

00-META/process/07-feature-branches.md

@@ -553,6 +553,13 @@ Every agent the mesh runs draws on one weekly allowance shared by every session,
 spends it many times faster. Up to five agents on the smaller model at once, at most two subagents on the
 larger one (the lead not counted).
 
+**A role started more than once has a definition.** When a role in the table above, or any other agent a
+playbook starts (a reviewer, a watcher, a survey), is started a second time, it is taken from a definition in
+the `.claude/agents/` of the repository whose work it serves, named in that repository's AGENTS.md and in the
+playbook step that starts it; the lead writes the definition the second time it writes such a prompt, in the
+same turn. Only what every session on a node or across the mesh needs is registered through the claude-code
+module (the operator, 2026-11-02). hq's words test reads `.claude/`, so a definition uses the glossary's words.
+
 ## Following and reporting
 
 Most lost time is waiting nobody saw. The lead therefore keeps a **wait list**: everything the feature waits