refactor(agent): plugin_llm/nous_rate_guard/moonshot_schema/provider_* — unify atomic write, compact docstrings, collapse defensive layers

agent/provider_projection.py

@@ -1,23 +1,13 @@
 """Fold an agent-as-provider's own activity back into Hermes' turn state.
 
-Some providers are *agents* (an ACP CLI behind a client shim; the codex
-app-server takes an analogous path in ``agent/codex_runtime.py``): they run
-their own tools inside their own session, so by the time Hermes sees the
-response that work is done. Those calls must never come back as pending
-``tool_calls`` (Hermes would re-run finished work), but two subsystems go blind
-if they are merely summarised into ``reasoning``:
-
-* the **self-improvement loop**, which replays ``messages`` to distil memories
-  and skills;
-* the **skill-review nudge**, whose ``_iters_since_skill`` counter only moves on
-  Hermes tool iterations.
-
-So the client hands both back on the completion object —
-``hermes_projected_messages`` (completed ``assistant(tool_calls=[…])`` +
-``tool(result)`` rows) and ``hermes_provider_tool_iterations`` — and this helper
-applies them. Ordinary OpenAI-compatible clients set neither and are unaffected.
-The splice is append-only through ``append_message`` so rows carry a timestamp
-and persist like any other live-transcript append.
+Agent providers (ACP CLI shims, the codex app-server) run their own tools, so that
+work must never come back as pending ``tool_calls`` (Hermes would re-run it) — but
+the self-improvement loop (replays ``messages``) and the skill-review nudge
+(``_iters_since_skill`` counter) go blind if it is merely summarised into
+``reasoning``. The client hands back ``hermes_projected_messages`` (completed
+assistant/tool rows) and ``hermes_provider_tool_iterations`` on the completion
+object; this helper applies them append-only via ``append_message`` (timestamped,
+persisted). Ordinary OpenAI-compatible clients set neither and are unaffected.
 """
 
 from __future__ import annotations
@@ -46,9 +36,7 @@ def splice_provider_projection(
         append_message(messages, row)
     if rows:
         logger.debug(
-            "spliced %d provider-projected transcript row(s) from %s",
-            len(rows),
-            getattr(agent, "provider", "?"),
+            "spliced %d provider-projected transcript row(s) from %s", len(rows), getattr(agent, "provider", "?"),
         )
 
     try: