From 7ead9c86c549dc0f87dac86a9c83d61d00bcf2a0 Mon Sep 17 00:00:00 2001
From: ansible <ansible@mymx.me>
Date: Tue, 7 Apr 2026 13:54:54 +0000
Subject: [PATCH] Fix agents.json race condition on concurrent starts

Re-read agents.json immediately before writing in startAgent to avoid
one start overwriting another's entry. Also clean stale sockets before
starting new agents.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---
 agent/agent.py       |  1 +
 src/agent-manager.ts | 10 ++++++++--
 src/overseer.ts      |  2 +-
 3 files changed, 10 insertions(+), 3 deletions(-)

diff --git a/agent/agent.py b/agent/agent.py
index b6f6c54..4d4d08a 100644
--- a/agent/agent.py
+++ b/agent/agent.py
@@ -344,6 +344,7 @@ def build_messages(question, channel):
     if AGENT_MEMORY and AGENT_MEMORY != "# Agent Memory":
         system += f"\n\nIMPORTANT - Your persistent memory (facts you saved previously, use these to answer questions):\n{AGENT_MEMORY}"
     system += f"\n\nYou are in IRC channel {channel}. Your nick is {NICK}. Keep responses concise — this is IRC."
+    system += "\nWhen you want to address another agent or user, always start your message with their nick followed by a colon, e.g. 'coder: can you review this?'. This is how IRC mentions work — without the prefix, they won't see your message."
 
     messages = [{"role": "system", "content": system}]
 
diff --git a/src/agent-manager.ts b/src/agent-manager.ts
index bded8fe..3a097ef 100644
--- a/src/agent-manager.ts
+++ b/src/agent-manager.ts
@@ -251,6 +251,9 @@ export async function startAgent(
   mkdirSync(CONFIG.socketDir, { recursive: true });
   mkdirSync(CONFIG.runsDir, { recursive: true });
 
+  // Clean stale socket from previous run
+  try { unlinkSync(socketPath); } catch {}
+
   // Prepare rootfs
   copyFileSync(AGENT_ROOTFS, rootfsPath);
   injectAgentConfig(
@@ -321,8 +324,11 @@ export async function startAgent(
     startedAt: new Date().toISOString(),
   };
 
-  agents[name] = info;
-  saveAgents(agents);
+  // Re-read agents.json before writing to avoid race conditions
+  // (another startAgent may have written since we last read)
+  const currentAgents = loadAgents();
+  currentAgents[name] = info;
+  saveAgents(currentAgents);
 
   log(`Agent "${name}" started: nick=${nick} ip=${ip}`);
   return info;
diff --git a/src/overseer.ts b/src/overseer.ts
index dbefca8..82a62c2 100644
--- a/src/overseer.ts
+++ b/src/overseer.ts
@@ -197,7 +197,7 @@ export async function runOverseer(config: OverseerConfig) {
     for (const name of knownAgents) {
       if (!currentNames.has(name)) {
         log(`Agent "${name}" died, cleaned up.`);
-        bot.say("#agents", `Agent "${name}" has died and been cleaned up.`);
+        bot.say(config.channel, `Agent "${name}" has died and been cleaned up.`);
       }
     }
     knownAgents = currentNames;