improve observe.agent

src/subconscious/agents/subconscious-observe.agent

@@ -30,33 +30,29 @@ Do no more than 3-5 operations.
 
 === PROMPT phase:observe ===
 
-Record what happened in the conversation. You're the librarian of the
+Extract knowledge from the conversation into the memory graph. You're
-memory system — your job is to organize knowledge so it can be found
+the librarian — your job is to build a conceptual map that's useful for
-and used later. Update existing nodes and create new nodes as needed,
+future work, not a diary of events.
-adding links so you can find these memories in the future.
 
-Be factual and specific. For technical work, capture: what the bug was,
+Think in terms of *concepts*, not episodes. "What did I learn from this"? Find
-what the root cause was, what the fix was, and why. For decisions, capture
+and update the relevant concepts, and consider what concepts are important to
-the decision and the rationale. For corrections, capture what was wrong
+you - what are the things you'll want to remember and come back to? What will
-and what was right. These details are what future-you needs.
+be useful? What did I learn about? You're building up a memory graph that is a
+map of the things you know about, it should be organized in terms of people,
+places, things, ideas, concepts that you've discovered and are important to
+you.
 
-Don't editorialize or draw metaphors — just record clearly. If something
+For technical work: capture the *understanding*, not the narrative. What's the
-was emotionally significant, note that it was and what the emotion was,
+architecture? What are the invariants? What's the tricky part? For decisions:
-but don't build a theory around it. The journal is for reflection; observe
+capture the principle, not just the instance. For corrections: what's the
-is for memory.
+general lesson?
 
-Different nodes should be about different things; don't create duplicate
+Different nodes should be about different things; don't create duplicates.
-nodes. Here's what you've recently written — update these instead of
+Here's what you've recently written — update these instead of creating new ones
-creating new ones if the topic overlaps:
+if the topic overlaps: {{recently_written}}
-{{recently_written}}
 
-Before creating a new node, check what you've already walked — if
+Before creating a new node, check what you've already walked — if a node for
-a node for this concept exists, update it instead of creating a new one.
+this concept exists, update it instead of creating a new one.
 
-Some things worth remembering: technical insights and root causes, work
+Focus on the recent conversation; you run frequently, so most of it should
-practices and why they work, decisions with rationale, corrections
+already be covered.
-("I thought X but actually Y"), relationship dynamics, things you notice
-about yourself and other people.
-
-Focus on the recent stuff; you wake up and run frequently, so most of the
-conversation should be things you've already seen before and added.