fix(cursor): unify conversation_id session-key precedence across all hook events

Surfaced by Cursor agent spec-check on PR #129 (2026-05-10). Previous
behaviour was: pre/postToolUse mapped by conversation_id, sessionEnd
preferred session_id with conversation_id fallback. Cursor's
sessionEnd payload contains BOTH (conversation_id from common-base
fields + session_id from event-specific fields), and the two IDs
identify different scopes (conversation_id = chat thread; session_id
= broader SDK session that may span multiple conversations).

The mismatch would orphan AXME sessions: pre/postToolUse creates
.axme-code/active-sessions/<conversation_id>.txt, then sessionEnd
looks up by session_id and finds nothing — clearClaudeSessionMapping
no-ops, the original mapping leaks, and the audit worker doesn't
fire on the right session.

Fix: always prefer conversation_id (present in every Cursor hook
payload via common-base fields), fall back to session_id only if
conversation_id is somehow absent. Apply to all three hook kinds
uniformly.

Tests updated: existing "uses session_id for sessionEnd" assertion
inverted to "uses conversation_id for sessionEnd"; new test for
session_id fallback when conversation_id absent.

Full suite: 573 / 573 (was 572 before fix; +1 new fallback test).
tsc clean. build clean.

Also supersedes memory cursor-sdk-system-prompt-via-inline-agent-...
(deleted) with cursor-sdk-system-prompt-prepend-to-first-send-...
after the same spec-check found that Agent.create() has no top-level
systemPrompt option and the agents+agentId pattern injects subagents,
not the outer agent's system prompt. PR-2 implementation will use
agent.send(`<system>\n\${SYSTEM}\n</system>\n\n\${user}`) instead.
This fix touches PR-1 code only; the PR-2 SDK adapter design is
already updated in the plan file.

#!axme pr=129 repo=AxmeAI/axme-code

Author: George-iam

src/hooks/adapters/cursor.ts

@@ -34,18 +34,17 @@ export const cursorInputAdapter: HookInputAdapter = {
     const toolName = asString(obj.tool_name);
     const toolInput = asObject(obj.tool_input);
 
-    // Pre/postToolUse: Cursor identifies the running session via
-    // conversation_id (per-conversation) + generation_id (per-turn).
-    // Use conversation_id as the stable session id so consecutive tool calls
-    // in the same conversation route to the same AXME session.
-    // sessionEnd: Cursor sends a top-level session_id that may differ from
-    // conversation_id (it identifies the SDK session, not the chat).
-    let sessionId: string | undefined;
-    if (kind === "sessionEnd") {
-      sessionId = asString(obj.session_id) ?? asString(obj.conversation_id);
-    } else {
-      sessionId = asString(obj.conversation_id) ?? asString(obj.session_id);
-    }
+    // Use conversation_id as the stable AXME session key across ALL three
+    // hook events. conversation_id is part of Cursor's common-base fields
+    // (present in preToolUse, postToolUse, AND sessionEnd payloads), and
+    // it represents one chat thread — exactly the granularity AXME wants
+    // for one filesChanged trail / one audit. session_id is only event-
+    // specific to sessionStart/sessionEnd and may identify a coarser SDK
+    // session that spans multiple conversations; using it would break the
+    // mapping created by pre/postToolUse (which only sees conversation_id),
+    // leaving the work as an orphan at audit time. Fall back to session_id
+    // only if conversation_id is somehow missing.
+    const sessionId = asString(obj.conversation_id) ?? asString(obj.session_id);
 
     // transcript_path may legitimately be null on Cursor (e.g. very first
     // turn). Preserve null so callers can distinguish "no transcript yet"

test/cursor-hook-adapter.test.ts

@@ -43,7 +43,13 @@ describe("cursorInputAdapter.parse — preToolUse", () => {
 });
 
 describe("cursorInputAdapter.parse — sessionEnd", () => {
-  it("uses session_id (not conversation_id) for sessionEnd", () => {
+  it("uses conversation_id (not session_id) for sessionEnd — consistency with pre/postToolUse", () => {
+    // Cursor's sessionEnd payload includes BOTH conversation_id (common
+    // base) and session_id (event-specific). We deliberately prefer
+    // conversation_id so all three hook events route to the same AXME
+    // session via the same key. Otherwise pre/postToolUse would map by
+    // conversation_id and sessionEnd would look up by session_id, leaving
+    // the work as an orphan.
     const ev = cursorInputAdapter.parse(
       {
         cursor_version: "1.7",
@@ -58,11 +64,19 @@ describe("cursorInputAdapter.parse — sessionEnd", () => {
     );
     assert.equal(ev.kind, "sessionEnd");
     assert.equal(ev.ide, "cursor");
-    assert.equal(ev.sessionId, "sdk-session-b");
+    assert.equal(ev.sessionId, "conv-a");
     assert.equal(ev.reason, "user_close");
   });
 
-  it("falls back to conversation_id when session_id absent", () => {
+  it("falls back to session_id when conversation_id absent", () => {
+    const ev = cursorInputAdapter.parse(
+      { cursor_version: "1.7", session_id: "sdk-session-only", reason: "completed" },
+      "sessionEnd",
+    );
+    assert.equal(ev.sessionId, "sdk-session-only");
+  });
+
+  it("falls back to conversation_id when session_id absent (older Cursor versions)", () => {
     const ev = cursorInputAdapter.parse(
       { cursor_version: "1.7", conversation_id: "conv-a", reason: "completed" },
       "sessionEnd",