coderouter-cli 2.8.1__py3-none-any.whl → 2.9.1__py3-none-any.whl

coderouter/doctor.py

@@ -1834,6 +1834,56 @@ async def _probe_cache(
     )
 
 
+# ---------------------------------------------------------------------------
+# Phase 2c migration guard (docs/designs/agent-cli-plugin-extraction.md §5.2,
+# §7 "2c" row): a config-level pre-check, not an HTTP probe. Core removed
+# its in-core agent_cli adapter in Phase 2c — kind="agent_cli" now resolves
+# ONLY via the coderouter-plugin-agents adapter plugin
+# (coderouter/adapters/registry.py's build_adapter). A provider that still
+# declares kind="agent_cli" without the plugin enabled will fail at engine
+# startup with a ValueError; surfacing it here lets `coderouter doctor`
+# catch it ahead of `coderouter serve`, with a copy-paste-able fix.
+# ---------------------------------------------------------------------------
+
+
+def _check_agent_cli_plugin_migration(
+    config: CodeRouterConfig, provider: ProviderConfig
+) -> ProbeResult | None:
+    """Warn when ``provider`` needs the agent_cli plugin but it isn't enabled.
+
+    Returns ``None`` (no finding) when ``provider.kind`` isn't
+    ``"agent_cli"``, or when ``"agents"`` is already listed in
+    ``plugins.enabled`` — i.e. nothing actionable to report.
+    """
+    if provider.kind != "agent_cli":
+        return None
+    enabled = config.plugins.enabled if config.plugins is not None else []
+    if "agents" in enabled:
+        return None
+    return ProbeResult(
+        name="agent-cli-plugin-migration",
+        verdict=ProbeVerdict.NEEDS_TUNING,
+        detail=(
+            f"provider {provider.name!r} declares kind='agent_cli', but the "
+            "coderouter-plugin-agents adapter plugin is not enabled. As of "
+            "v2.9.0 (Phase 2c), Core no longer ships an in-core agent_cli "
+            "adapter — this provider cannot serve requests until the "
+            "plugin is installed and listed in plugins.enabled."
+        ),
+        target_file="providers.yaml",
+        suggested_patch=(
+            "# install the plugin:\n"
+            'pip install "coderouter-plugin-agents @ '
+            'git+https://github.com/zephel01/coderouter-plugin-agents"\n'
+            "\n"
+            "# providers.yaml — enable it (merge into any existing list):\n"
+            "plugins:\n"
+            "  enabled:\n"
+            "    - agents\n"
+        ),
+    )
+
+
 # ---------------------------------------------------------------------------
 # Orchestration
 # ---------------------------------------------------------------------------
@@ -1872,6 +1922,14 @@ async def check_model(
         resolved_caps=resolved,
     )
 
+    migration = _check_agent_cli_plugin_migration(config, provider)
+    if migration is not None:
+        # Nothing to probe over HTTP — the provider has no adapter until
+        # the plugin is enabled. Report the finding and stop here rather
+        # than sending HTTP probes at a (kind-wise meaningless) base_url.
+        report.results.append(migration)
+        return report
+
     auth_result = await _probe_auth_and_basic_chat(provider)
     report.results.append(auth_result)
 

coderouter/ingress/app.py

@@ -305,6 +305,13 @@ def create_app(config_path: str | None = None) -> FastAPI:
                 },
             )
 
+        # launcher-model-swap.md §6.6 known-trap #9: the TTL sweeper is a
+        # background task like continuous-probe above — start it once the
+        # event loop is actually running, not at app-construction time.
+        swap_manager = getattr(app.state, "swap", None)
+        if swap_manager is not None:
+            await swap_manager.start()
+
         yield
 
         # Graceful shutdown of probe task
@@ -313,6 +320,14 @@ def create_app(config_path: str | None = None) -> FastAPI:
             with contextlib.suppress(Exception):
                 await probe_task
 
+        # Cancel the swap TTL sweeper *before* shutdown_launcher tears down
+        # every ManagedProcess below — shutdown_launcher already stops
+        # swap-spawned processes too (they're in the same registry), so
+        # this only needs to stop the sweeper loop itself, not race it.
+        if swap_manager is not None:
+            with contextlib.suppress(Exception):
+                await swap_manager.stop()
+
         # Launcher: stop child llama.cpp / vllm processes so they don't orphan.
         from coderouter.ingress.launcher_routes import shutdown_launcher
 
@@ -371,6 +386,20 @@ def create_app(config_path: str | None = None) -> FastAPI:
     app.state.engine = engine
     app.state.config = config
 
+    # Phase 1 on-demand model swap (docs/designs/launcher-model-swap.md).
+    # None (default) leaves the engine's swap dispatch hooks as cheap
+    # no-ops — zero behavior change for deployments that don't opt in.
+    # Constructed here (needs the real ``app`` for app.state.launcher /
+    # app.state.engine) but the TTL sweeper task itself is only started
+    # from the lifespan below, once the event loop is actually running.
+    swap_cfg = config.launcher.swap if config.launcher is not None else None
+    if swap_cfg is not None and swap_cfg.enabled:
+        from coderouter.launcher_swap import SwapManager
+
+        swap_manager = SwapManager(app, swap_cfg, config.launcher)
+        app.state.swap = swap_manager
+        engine.attach_swap_manager(swap_manager)
+
     # H8: DNS-rebinding protection. Applied to every route so a hostname an
     # attacker controls cannot be pointed at loopback and used to drive the
     # local API from a victim's browser. Extra hostnames for deliberate