merge: memory tool guidance prompt

crates/pod/src/prompt/catalog.rs

@@ -507,6 +507,18 @@ mod tests {
         assert!(rendered.contains("mark_read_required"));
     }
 
+    #[test]
+    fn internal_worker_prompts_do_not_include_default_memory_guidance() {
+        let cat = PromptCatalog::builtins_only().unwrap();
+        let compact = cat.compact_system().unwrap();
+        let extract = cat.memory_extract_system("Japanese").unwrap();
+        let consolidate = cat.memory_consolidation_system("Japanese").unwrap();
+        for rendered in [compact, extract, consolidate] {
+            assert!(!rendered.contains("### Memory and knowledge"));
+            assert!(!rendered.contains("Do not query memory every turn"));
+        }
+    }
+
     #[test]
     fn memory_worker_prompts_include_language() {
         let cat = PromptCatalog::builtins_only().unwrap();

crates/pod/src/prompt/system.rs

@@ -440,7 +440,11 @@ mod tests {
         let rendered = tmpl
             .render(&ctx(dir.path(), &scope, vec!["Read".into()], None))
             .unwrap();
-        // Builtin default body must expose the language policy.
+        // Builtin default body must expose the tool and language policies.
+        assert!(rendered.contains("### Memory and knowledge"));
+        assert!(rendered.contains("MemoryQuery"));
+        assert!(rendered.contains("MemoryRead(kind=summary)"));
+        assert!(rendered.contains("Do not query memory every turn"));
         assert!(rendered.contains("## Language"));
         assert!(rendered.contains("`language`: `match the user's language"));
         // Trailing section must be present.

resources/prompts/common/tool-usage.md

@@ -5,3 +5,10 @@ When searching, use grep/glob primitives rather than shell pipelines.
 
 You can run multiple tools simultaneously by calling them within a single response.
 It is recommended to run tools that handle asynchronous processing, such as queries and readings, in batches.
+
+### Memory and knowledge
+
+For past decisions, prior requests, durable preferences, project history, or why something was done, use targeted lookup instead of guessing from vague recollection.
+Use `MemoryQuery` for durable memory records (summary, decisions, requests), `KnowledgeQuery` for project knowledge, `MemoryRead(kind=summary)` for the full memory summary, and `MemoryRead` on returned slugs when excerpts are insufficient.
+Resident memory and knowledge are helpful context but may be stale; current user instructions, repository files, tickets, git history, and session logs are authoritative for exact current state.
+Do not query memory every turn, and normally prefer read/query tools; use `MemoryWrite`, `MemoryEdit`, or `MemoryDelete` only when explicitly asked or in a memory maintenance worker.