diff --git a/agentburn/mcp.py b/agentburn/mcp.py
index e4d2bc5..08065b7 100644
--- a/agentburn/mcp.py
+++ b/agentburn/mcp.py
@@ -28,7 +28,7 @@
     "properties": {
         "agent": {
             "type": "string",
-            "description": "hermes | openclaw | claude-code (default: first detected)",
+            "description": "hermes | openclaw | claude-code | codex | gemini | opencode (default: first detected on this machine)",
         },
         "days": {"type": "integer", "description": "window in days (default 30, 0 = all time)"},
         "source": {
@@ -87,7 +87,11 @@
     },
     {
         "name": "burn_card",
-        "description": "Anonymized shareable burn summary (plain text, safe to post).",
+        "description": (
+            "Anonymized shareable burn summary as plain text: categories and counters only, "
+            "no session titles or file paths, safe to post publicly. Use it when the user wants "
+            "to share their numbers; for analysis use burn_report or burn_why instead."
+        ),
         "inputSchema": WINDOW,
     },
 ]
diff --git a/skill/agentburn/SKILL.md b/skill/agentburn/SKILL.md
index 2004e5e..22b49dc 100644
--- a/skill/agentburn/SKILL.md
+++ b/skill/agentburn/SKILL.md
@@ -45,8 +45,8 @@ database read-only. Use it instead of guessing about costs.
 5. For one channel ("what did you do in telegram?"):
    add `--source telegram` (or cron / heartbeat / subagent / cli).
 6. Answer in the user's language, lead with the verdict
-   (pace + dominant source), quote at most 3 numbers, then the single
-   highest-impact change. Mark estimated costs with "~".
+   (pace + dominant source), cite only the numbers the verdict rests on,
+   then the single highest-impact change. Mark estimated costs with "~".
 
 ## Fixing things