Fix LiteLLM context budgets and direct computer use

Keep efficient context windows separate from expanded maxima throughout discovery, display and provider budgets. Expose correlated computer-use requests and actions with existing permissions, preserve screenshots as images, and retain them across checkpoints, compaction and explicit history recall.

Validation: cargo build on Linux passed. Added focused regression coverage; test execution and native GUI verification remain outside this CLI build-only workflow.

app/src/ai/provider/response_translator.rs

@@ -839,6 +839,14 @@ pub fn build_tool_call_message(
         serde_json::from_str(tool_input_json).unwrap_or(serde_json::json!({}));
 
     let tool = match tool_name {
+        "request_computer_use" | "use_computer" => {
+            crate::ai::runtime::action_from_tool_call(
+                task_id,
+                &galaxy_agent_core::ToolCall { id: tool_use_id.to_string(), name: tool_name.to_string(), arguments: input.clone() },
+                &ai::skills::SkillPathOrigin::Local,
+                &std::collections::HashMap::new(),
+            ).ok().and_then(|action| crate::ai::agent::provider_history::computer_use_tool(&action.action))
+        }
         "run_shell_command" => {
             let command = input
                 .get("command")
@@ -1596,6 +1604,8 @@ fn run_agents_harness_from_str(harness_type: &str) -> Option<api::Harness> {
 
 /// Built-in tools that Galaxy knows how to execute directly.
 const KNOWN_TOOLS: &[&str] = &[
+    "request_computer_use",
+    "use_computer",
     "run_shell_command",
     "read_files",
     "apply_file_diffs",