chore(prompts): stop generating verb tables for driver-based roles

regenerate_verb_tables.py looped every role in ROLE_CONFIGS, emitting a
_generated/<role>.md for prompter and secretary too. Both intentionally keep
only note+evidence in role_config — their real tools live in their agent_sdk
drivers (intake: propose_draft; secretary: read_state/read_task/
submit_directive, the last gated through the backend /directives), and neither
uses the _generated/<role>.md prompt-composition path. So the generated tables
understated those roles and showed up as perpetually-untracked noise.

Skip the driver-based roles (_DRIVER_BASED_ROLES) in both the aggregate verbs.md
and the per-role file output, with a comment pointing at the real surfaces.
Regenerated verbs.md drops the two misleading sections.

agents/prompts/_generated/verbs.md

@@ -7,6 +7,9 @@ Run `uv run python scripts/regenerate_verb_tables.py` after changing
 any role config or schema. Role prompts reference this file's sections
 as the source of truth for verb signatures.
 
+Driver-based roles (prompter, secretary) are intentionally omitted — their
+real tools live in their agent_sdk drivers, not role_config.
+
 ## developer
 
 ### Flow verbs
@@ -246,33 +249,3 @@ as the source of truth for verb signatures.
 | `notify_get` | `notify_get(notification_id: UUID)` |
 | `channels` | `channels()` |
 
-## prompter
-
-### Flow verbs
-
-| Verb | Body schema |
-|------|-------------|
-| `i_am_idle` | `i_am_idle()` |
-
-### Content (do) tools
-
-| Tool | Body schema |
-|------|-------------|
-| `note` | `note(text: str, scope: str = 'note', task_id: UUID | None = None, title: str | None = None, context: str = '', options: list[str | str] | None = None, chosen: str = '', rationale: str = '', consequences: list[str] | None = None, what_done: str = '', what_learned: str = '', what_struggled: str = '', next_steps: list[str] | None = None)` |
-| `evidence` | `evidence(task_id: UUID)` |
-
-## secretary
-
-### Flow verbs
-
-| Verb | Body schema |
-|------|-------------|
-| `i_am_idle` | `i_am_idle()` |
-
-### Content (do) tools
-
-| Tool | Body schema |
-|------|-------------|
-| `note` | `note(text: str, scope: str = 'note', task_id: UUID | None = None, title: str | None = None, context: str = '', options: list[str | str] | None = None, chosen: str = '', rationale: str = '', consequences: list[str] | None = None, what_done: str = '', what_learned: str = '', what_struggled: str = '', next_steps: list[str] | None = None)` |
-| `evidence` | `evidence(task_id: UUID)` |
-

scripts/regenerate_verb_tables.py

@@ -166,6 +166,18 @@ def _render_role_section(role: str) -> str:
     return "\n".join(lines)
 
 
+# Roles whose real tool surface lives in a dedicated agent_sdk driver, NOT in
+# role_config's flow_tools/do_tools — so a table rendered from role_config would
+# UNDERSTATE them and mislead. The prompter (intake) registers `propose_draft`
+# (+ read-only Read/Grep/Glob/Task) in `agent_sdk/intake_driver.py`; the
+# secretary registers `read_state`/`read_task`/`submit_directive` in
+# `agent_sdk/secretary_driver.py`. Both keep only `note`+`evidence` in
+# role_config by design, and neither uses the `_generated/<role>.md` prompt-
+# composition path (they run from their own drivers). Skip them so this script
+# never emits a thin, misleading table for them.
+_DRIVER_BASED_ROLES: frozenset[str] = frozenset({"prompter", "secretary"})
+
+
 def main() -> None:
     sections = [
         "<!-- AUTOGENERATED by scripts/regenerate_verb_tables.py. -->",
@@ -177,8 +189,13 @@ def main() -> None:
         "any role config or schema. Role prompts reference this file's sections",
         "as the source of truth for verb signatures.",
         "",
+        "Driver-based roles (prompter, secretary) are intentionally omitted — their",
+        "real tools live in their agent_sdk drivers, not role_config.",
+        "",
     ]
     for role in ROLE_CONFIGS:
+        if role in _DRIVER_BASED_ROLES:
+            continue
         sections.append(_render_role_section(role))
 
     _OUT_DIR.mkdir(parents=True, exist_ok=True)
@@ -187,7 +204,11 @@ def main() -> None:
 
     # Also write per-role files so the prompt composer can include each
     # role's verb table directly without parsing a multi-section document.
+    # Driver-based roles are skipped (see _DRIVER_BASED_ROLES) — they don't use
+    # this composition path and a role_config-only table would understate them.
     for role in ROLE_CONFIGS:
+        if role in _DRIVER_BASED_ROLES:
+            continue
         per_role = _OUT_DIR / f"{role}.md"
         header = (
             "<!-- AUTOGENERATED by scripts/regenerate_verb_tables.py. -->\n"