chron-mcp 0.1.6 → 0.1.7

package/dist/server.js

@@ -9,7 +9,7 @@ const roleEnum = zod_1.z.enum(['user', 'assistant']);
 function createServer(db) {
     const server = new mcp_js_1.McpServer({
         name: 'chron',
-        version: '0.1.6',
+        version: '0.1.7',
     });
     // @ts-expect-error — MCP SDK deep generic type instantiation limit hit; runtime types are correct
     server.tool('start_session', 'Create a new audit session or resume an existing one by title. Call this at the start of every conversation.', {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "chron-mcp",
-  "version": "0.1.6",
+  "version": "0.1.7",
   "mcpName": "io.github.SirinivasK/chron",
   "description": "Audit-grade timestamped logs for every AI conversation",
   "repository": {

package/skills/chron.skill.md

@@ -16,9 +16,9 @@ Call `get_session_history` for the returned `session_id`. If prior messages exis
 The rule is simple: **any time you output text and wait for input, log it.** This includes mid-task check-ins, not just clean user→assistant→user flows.
 
 1. Call `log_message(session_id, role="user", content=<exact user message>)` → returns T1
-2. Compose your response body (plain text only — no timestamp header)
+2. Compose your response body internally (plain text only — no timestamp header). **Do NOT output any text yet.**
 3. Call `log_message(session_id, role="assistant", content=<response body>)` → returns T2
-4. Output header + body:
+4. Only now output the header + body — exactly once:
 
 ```
 [user: YYYY-MM-DD HH:MM:SS ±HH:MM | assistant: YYYY-MM-DD HH:MM:SS ±HH:MM]
@@ -26,6 +26,8 @@ The rule is simple: **any time you output text and wait for input, log it.** Thi
 <response body>
 ```
 
+**Never output text before step 3.** Doing so causes the response to appear twice — once before the log call and once after. Compose fully in your head, log, then output.
+
 T1 and T2 are real server timestamps — T2 − T1 is the actual generation time. The DB stores clean body content without the header.
 
 **Mid-task check-ins follow the same rule.** If you pause mid-task and ask the user something ("I've done the analysis, what's next?" / "Which approach do you prefer?"), that pause is an exchange — log the assistant message before outputting it, log the user's reply when it arrives. Every turn that crosses the user boundary gets timestamped, regardless of whether the task is "complete."