aboutsummaryrefslogtreecommitdiff
path: root/help.html
diff options
context:
space:
mode:
Diffstat (limited to 'help.html')
-rw-r--r--help.html17
1 files changed, 17 insertions, 0 deletions
diff --git a/help.html b/help.html
index acffae6..5bf3543 100644
--- a/help.html
+++ b/help.html
@@ -1689,6 +1689,23 @@ listmods</pre>
<h2>📋 Changelog</h2>
<div class="version-item">
+ <div class="v-header"><span class="v-num">v4.6.83–84 — October 2026</span>
+ <span class="v-date">world_model_query</span></div>
+ <ul>
+ <li><strong>The first tool the system prompt tells the model to call no longer fails.</strong>
+ <code>world_model_query</code> with the default detail parsed
+ <code>_modelDigest()</code> as JSON — but that digest was a JSON <em>string</em> cut at 4000
+ characters with <code>…(digest truncated)</code> glued on, so it threw
+ <code>Unexpected end of JSON input</code> every time. <code>detail:"full"</code> worked only
+ because it returned the object and never parsed anything.</li>
+ <li><strong>A digest shrinks, it does not break.</strong> The model now builds a real
+ snapshot object and drops list lengths (40 → 20 → 10 → 5 → 3 → 1 → 0), and if still too
+ large drops whole sections, reporting what it left out in <code>omitted</code>. The string
+ form is only produced at the edge, and is always parseable.</li>
+ </ul>
+ </div>
+
+ <div class="version-item">
<div class="v-header"><span class="v-num">v4.6.82 — October 2026</span>
<span class="v-date">skill-create parser</span></div>
<ul>