runner: call memory_search directly instead of spawning poc-hook

The agent was shelling out to poc-hook which shells out to memory-search.
Now that everything is one crate, just call the library function. Removes
subprocess overhead on every user message.

Co-Authored-By: Proof of Concept <poc@bcachefs.org>

src/agent/runner.rs

@@ -17,7 +17,6 @@ use anyhow::Result;
 use tiktoken_rs::CoreBPE;
 
 use std::io::Write;
-use std::process::{Command, Stdio};
 
 use crate::agent::api::ApiClient;
 use crate::agent::journal;
@@ -129,8 +128,9 @@ impl Agent {
         agent
     }
 
-    /// Run poc-hook for a given event, returning any output to inject.
+    /// Run memory search for a given event, returning any output to inject.
-    fn run_hook(&self, event: &str, prompt: &str) -> Option<String> {
+    /// Direct library call — no subprocess needed since everything is one crate.
+    fn run_hook(&self, event: &str, _prompt: &str) -> Option<String> {
         let transcript_path = self.conversation_log.as_ref()
             .map(|l| l.path().to_string_lossy().to_string())
             .unwrap_or_default();
@@ -139,23 +139,9 @@ impl Agent {
             "hook_event_name": event,
             "session_id": self.session_id,
             "transcript_path": transcript_path,
-            "prompt": prompt,
         });
 
-        let mut child = Command::new("poc-hook")
+        let text = crate::memory_search::run_hook(&hook_input.to_string());
-            .stdin(Stdio::piped())
-            .stdout(Stdio::piped())
-            .stderr(Stdio::null())
-            .spawn()
-            .ok()?;
-
-        if let Some(ref mut stdin) = child.stdin {
-            let _ = stdin.write_all(hook_input.to_string().as_bytes());
-        }
-        drop(child.stdin.take());
-
-        let output = child.wait_with_output().ok()?;
-        let text = String::from_utf8_lossy(&output.stdout).to_string();
         if text.trim().is_empty() {
             None
         } else {