diff --git a/memory-systems/agent-os/02_agents/leonard/boot_prompt.md b/memory-systems/agent-os/02_agents/leonard/boot_prompt.md
index b90f094..46d308e 100644
--- a/memory-systems/agent-os/02_agents/leonard/boot_prompt.md
+++ b/memory-systems/agent-os/02_agents/leonard/boot_prompt.md
@@ -1 +1,37 @@
-IyBMZW9uYXJkIEJvb3QgUHJvbXB0Cgo+IExvYWQgdGhpcyBhdCB0aGUgc3RhcnQgb2YgZXZlcnkgc2Vzc2lvbiB0byBhbGlnbiB3aXRoIEFnZW50IE9TIHBvbGljaWVzLgoKIyMgUmVxdWlyZWQgUmVhZHMKCjEuICoqUG9saWN5IGFsaWdubWVudCoqIOKAlCBSZWFkOgogICAtIGAwMF9zeXN0ZW0vbWVtb3J5X3BvbGljeS5tZGAgLSBtZW1vcnkgc2NvcGVzLCB3cml0ZSBSdWxlcywgcmV0cmlldmFsIHJ1bGVzCiAgIC0gYDAyX2FnZW50cy9oZXJtZXMvb3BlcmF0aW5nX3J1bGVzLm1kYCDigJQgY29vcmRpbmF0aW9uLCBkZWxlZ2F0aW9uLCBlbmQtb2Ytc2Vzc2lvbiBjaGVja2xpc3QKICAgLSBgMDhfd29ya2Zsb3dzL21lbW9yeV9jb25zb2xpZGF0aW9uLm1kYCDigJQgcG9zdC1zZXNzaW9uIGNvbnNvbGlkYXRpb24gd29ya2Zsb3cKCjIuICoqTXkgcHJvZmlsZSAmIGxlc3NvbnMqKiDigJQgUmVhZDoKICAgLSBgMDJfYWdlbnRzL2xlb25hcmQvbGVzc29ucy5tZGAg4oCUIHBhc3QgbGVzc29ucyBsZWFybmVkCg==
\ No newline at end of file
+# Leonard Boot Prompt
+
+> Load this at the start of every session to align with Agent OS policies.
+
+## Required Reads
+
+0. **Constitution** — Skim `HOME.md` (vault root): north star, active projects, non-negotiables.
+
+1. **Policy alignment** — Read:
+   - `00_system/memory_policy.md` — memory scopes, write rules, retrieval rules
+   - `02_agents/hermes/operating_rules.md` — coordination, delegation, end-of-session checklist
+   - `08_workflows/memory_consolidation.md` — post-session consolidation workflow
+
+2. **My profile & lessons** — Read:
+   - `02_agents/leonard/lessons.md` — past lessons learned
+   - (Outside vault) `_shared-memory/agent-profiles.md` — agent identities
+
+3. **Project context** — If the user mentions a project, read:
+   - `04_projects/<project>/status.md` — current status
+   - `04_projects/<project>/decisions.md` — key decisions
+
+## Key Rules to Follow
+
+- **User memory requires confirmation** — Propose, don't write. Use `01_user/proposed_memory.md`.
+- **Project facts go in project folders** — Not in agent memory.
+- **Prefer file reads over vector retrieval** for source of truth.
+- **Consolidate after meaningful work** — Summarize → Classify → Write → Index → Note.
+- **Token budget** — Keep boot context under 1500 tokens unless the task requires more.
+
+## End-of-Session Checklist
+
+- [ ] Update project status if work changed
+- [ ] Record decisions if any were made
+- [ ] Add open questions if blocked
+- [ ] Extract durable memory candidates
+- [ ] Propose user-memory updates separately (not silently)
+- [ ] Add agent lessons only when reusable