huskies: merge 1207 story fleet_resources MCP tool: on-demand host + per-container disk, load, and CPU/mem

server/src/http/gateway/mcp.rs

@@ -29,6 +29,8 @@ const GATEWAY_TOOLS: &[&str] = &[
     "prompt_permission",
     // One-shot container rebuild: build fresh image, swap container, preserve state.
     "project_rebuild",
+    // On-demand host + per-container disk/load/CPU/mem snapshot (story 1207).
+    "fleet_resources",
     // Read sled identity pins vs. live signed identity, and TOFU re-pin.
     "fleet_identity",
 ];
@@ -156,6 +158,31 @@ pub(crate) fn gateway_tool_definitions() -> Vec<Value> {
                 "required": ["name"]
             }
         }),
+        json!({
+            "name": "fleet_resources",
+            "description": "On-demand host and per-container resource snapshot: host disk free/total, load averages, core count, and memory; per-container CPU%/mem sourced via `docker stats`; and per-project `target/`+`.huskies/worktrees/` directory sizes sourced via `docker exec ... find` (bounded, skips node_modules/.git, cached briefly per container+path). Flags disk/load conditions past thresholds first so problems lead the response.",
+            "inputSchema": {
+                "type": "object",
+                "properties": {
+                    "disk_warn_gb": {
+                        "type": "integer",
+                        "description": "Host free-disk warn threshold in GB (default: 50)."
+                    },
+                    "disk_critical_gb": {
+                        "type": "integer",
+                        "description": "Host free-disk critical threshold in GB (default: 20)."
+                    },
+                    "load_warn_per_core": {
+                        "type": "number",
+                        "description": "1-minute load average per core above which a warn flag fires (default: 1.0)."
+                    },
+                    "load_critical_per_core": {
+                        "type": "number",
+                        "description": "1-minute load average per core above which a critical flag fires (default: 2.0)."
+                    }
+                }
+            }
+        }),
         json!({
             "name": "fleet_identity",
             "description": "Read mode (default): for every registered sled, report project, url, connected, the recorded pin (expected_node_id), the live signature-verified node_id from a signed challenge-response (never the unsigned /identity display field), and whether they match. Repin mode: capture a sled's live verified identity via TOFU and persist it as the new pin, refusing when the signature is missing or does not verify.",
@@ -440,6 +467,7 @@ async fn handle_gateway_tool(
         "agents.list" => handle_agents_list_tool(id),
         "prompt_permission" => handle_prompt_permission_tool(params, state, id).await,
         "project_rebuild" => handle_project_rebuild_tool(params, state, id).await,
+        "fleet_resources" => handle_fleet_resources_tool(params, state, id).await,
         "fleet_identity" => handle_fleet_identity_tool(params, state, id).await,
         _ => JsonRpcResponse::error(id, -32601, format!("Unknown gateway tool: {tool_name}")),
     }
@@ -943,6 +971,65 @@ async fn handle_project_rebuild_tool(
     )
 }
 
+/// Handle the `fleet_resources` gateway tool (story 1207).
+///
+/// Collects host disk/load/cpu/mem, per-container CPU%/mem (via `docker
+/// stats`), and per-project `target/`/`worktrees/` sizes (via `docker exec
+/// ... find`) — all sourced gateway-side rather than requiring each project's
+/// sled to self-report. Thresholds for the leading `flags` list are supplied
+/// as optional arguments (see `gateway_tool_definitions`), defaulting to
+/// `ResourceThresholds::default()`.
+async fn handle_fleet_resources_tool(
+    params: &Value,
+    state: &GatewayState,
+    id: Option<Value>,
+) -> JsonRpcResponse {
+    use crate::service::gateway::resources::ResourceThresholds;
+
+    let args = params.get("arguments").unwrap_or(params);
+    let defaults = ResourceThresholds::default();
+    let thresholds = ResourceThresholds {
+        disk_warn_gb: args
+            .get("disk_warn_gb")
+            .and_then(|v| v.as_u64())
+            .unwrap_or(defaults.disk_warn_gb),
+        disk_critical_gb: args
+            .get("disk_critical_gb")
+            .and_then(|v| v.as_u64())
+            .unwrap_or(defaults.disk_critical_gb),
+        load_warn_per_core: args
+            .get("load_warn_per_core")
+            .and_then(|v| v.as_f64())
+            .unwrap_or(defaults.load_warn_per_core),
+        load_critical_per_core: args
+            .get("load_critical_per_core")
+            .and_then(|v| v.as_f64())
+            .unwrap_or(defaults.load_critical_per_core),
+    };
+
+    let project_names: Vec<String> = state.projects.read().await.keys().cloned().collect();
+
+    match crate::service::gateway::resources::io::collect_fleet_resources(
+        &state.config_dir,
+        &project_names,
+        &thresholds,
+    )
+    .await
+    {
+        Ok(resources) => JsonRpcResponse::success(
+            id,
+            json!({
+                "content": [{
+                    "type": "text",
+                    "text": serde_json::to_string_pretty(&resources).unwrap_or_default()
+                }],
+                "resources": serde_json::to_value(&resources).unwrap_or(json!(null)),
+            }),
+        ),
+        Err(e) => JsonRpcResponse::error(id, -32603, format!("fleet_resources failed: {e}")),
+    }
+}
+
 /// Handle the `fleet_identity` gateway tool.
 ///
 /// Dispatches on the `action` argument: `"read"` (default) reports every