chore: promote staging to staging-promote/764e5867-24277535589 (2026-04-11 09:13 UTC)

.github/labeler.yml

@@ -1,5 +1,5 @@
-# Scope labels for actions/labeler@v6
-# Maps file path globs to scope labels. Multiple labels can apply per PR.
+# Labels for actions/labeler@v6
+# Maps file path globs to labels. Multiple labels can apply per PR.
 
 "scope: agent":
   - changed-files:
@@ -164,3 +164,9 @@
       - any-glob-to-any-file:
           - Cargo.toml
           - Cargo.lock
+
+"DB MIGRATION":
+  - changed-files:
+      - any-glob-to-any-file:
+          - migrations/**
+          - src/db/libsql_migrations.rs

.github/scripts/create-labels.sh

@@ -62,6 +62,9 @@ create "scope: ci"            "546E7A" "CI/CD workflows"
 create "scope: docs"          "78909C" "Documentation"
 create "scope: dependencies"  "90A4AE" "Dependency updates"
 
+echo "==> Creating coordination labels..."
+create "DB MIGRATION"         "C62828" "PR adds or modifies PostgreSQL or libSQL migration definitions"
+
 echo "==> Creating workflow labels..."
 create "skip-regression-check" "9E9E9E" "Acknowledged: fix without regression test"
 

.github/workflows/pr-label-scope.yml

@@ -6,12 +6,19 @@ on:
 
 permissions:
   contents: read
+  issues: write
   pull-requests: write
 
 jobs:
   scope:
     runs-on: ubuntu-latest
     steps:
+      - name: Ensure DB MIGRATION label exists
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          REPO: ${{ github.repository }}
+        run: gh label create "DB MIGRATION" --repo "$REPO" --color C62828 --description "PR adds or modifies PostgreSQL or libSQL migration definitions" --force
+
       - uses: actions/labeler@8558fd74291d67161a8a78ce36a881fa63b766a9 # v5
         with:
           configuration-path: .github/labeler.yml

.github/workflows/release.yml

@@ -470,6 +470,17 @@ jobs:
           # shellcheck disable=SC2086  # PRERELEASE_FLAG is '--prerelease' or empty
           gh release create "$RELEASE_TAG" --target "$RELEASE_COMMIT" $PRERELEASE_FLAG --title "$ANNOUNCEMENT_TITLE" --notes-file "$RUNNER_TEMP/notes.txt" artifacts/*
 
+  # Build and push Docker Hub images (:version, :latest, :sha-*) after the GitHub Release exists.
+  docker-image:
+    needs: host
+    if: ${{ always() && needs.host.result == 'success' }}
+    permissions:
+      contents: read
+      packages: read
+      actions: write
+    uses: ./.github/workflows/docker.yml
+    secrets: inherit
+
   # Commit patched manifest SHA256 checksums back to main so the repo
   # stays in sync with the released artifacts.
   update-registry-checksums:

channels-src/telegram/src/lib.rs

@@ -376,6 +376,26 @@ const TELEGRAM_STATUS_MAX_CHARS: usize = 600;
 /// Telegram's hard limit for message text length.
 const TELEGRAM_MAX_MESSAGE_LEN: usize = 4096;
 
+fn utf16_code_unit_len(text: &str) -> usize {
+    text.encode_utf16().count()
+}
+
+fn prefix_within_utf16_limit(text: &str, max_units: usize) -> usize {
+    let mut units = 0;
+    let mut end = 0;
+
+    for (byte_idx, ch) in text.char_indices() {
+        let ch_units = ch.len_utf16();
+        if units + ch_units > max_units {
+            break;
+        }
+        units += ch_units;
+        end = byte_idx + ch.len_utf8();
+    }
+
+    end
+}
+
 fn truncate_status_message(input: &str, max_chars: usize) -> String {
     let mut iter = input.chars();
     let truncated: String = iter.by_ref().take(max_chars).collect();
@@ -386,7 +406,7 @@ fn truncate_status_message(input: &str, max_chars: usize) -> String {
     }
 }
 
-/// Split a long message into chunks that fit within Telegram's 4096-char limit.
+/// Split a long message into chunks that fit within Telegram's 4096 UTF-16-unit limit.
 ///
 /// Tries to split at the most natural boundary available (in priority order):
 /// 1. Double newline (paragraph break)
@@ -395,28 +415,36 @@ fn truncate_status_message(input: &str, max_chars: usize) -> String {
 /// 4. Word boundary (space)
 /// 5. Hard cut at the limit (last resort for pathological input)
 fn split_message(text: &str) -> Vec<String> {
-    if text.chars().count() <= TELEGRAM_MAX_MESSAGE_LEN {
+    if utf16_code_unit_len(text) <= TELEGRAM_MAX_MESSAGE_LEN {
         return vec![text.to_string()];
     }
 
     let mut chunks: Vec<String> = Vec::new();
     let mut remaining = text;
 
     while !remaining.is_empty() {
-        // Count chars to find the byte offset for our window.
-        let window_bytes = remaining
-            .char_indices()
-            .take(TELEGRAM_MAX_MESSAGE_LEN)
-            .last()
-            .map(|(byte_idx, ch)| byte_idx + ch.len_utf8())
-            .unwrap_or(remaining.len());
+        // Find the longest UTF-8 prefix that fits within Telegram's UTF-16 limit.
+        let window_bytes = prefix_within_utf16_limit(remaining, TELEGRAM_MAX_MESSAGE_LEN);
 
         if window_bytes >= remaining.len() {
             // Remainder fits entirely.
             chunks.push(remaining.to_string());
             break;
         }
 
+        if window_bytes == 0 {
+            // Defensive fallback: make progress even if a future caller uses a
+            // smaller limit than a single scalar value can fit within.
+            let first_char_len = remaining
+                .chars()
+                .next()
+                .map(|ch| ch.len_utf8())
+                .unwrap_or(remaining.len());
+            chunks.push(remaining[..first_char_len].to_string());
+            remaining = &remaining[first_char_len..];
+            continue;
+        }
+
         let window = &remaining[..window_bytes];
 
         // 1. Double newline — best paragraph boundary
@@ -2268,6 +2296,10 @@ export!(TelegramChannel);
 mod tests {
     use super::*;
 
+    fn utf16_len(text: &str) -> usize {
+        text.encode_utf16().count()
+    }
+
     #[test]
     fn test_split_message_short() {
         let text = "Hello, world!";
@@ -2295,7 +2327,7 @@ mod tests {
         let chunks = split_message(&text);
         assert!(chunks.len() > 1, "expected multiple chunks");
         for chunk in &chunks {
-            assert!(chunk.chars().count() <= TELEGRAM_MAX_MESSAGE_LEN);
+            assert!(utf16_len(chunk) <= TELEGRAM_MAX_MESSAGE_LEN);
         }
         // Rejoined chunks must equal the original text exactly.
         let rejoined = chunks.join(" ");
@@ -2311,7 +2343,7 @@ mod tests {
         assert!(text.len() > TELEGRAM_MAX_MESSAGE_LEN);
         let chunks = split_message(&text);
         for chunk in &chunks {
-            assert!(chunk.chars().count() <= TELEGRAM_MAX_MESSAGE_LEN);
+            assert!(utf16_len(chunk) <= TELEGRAM_MAX_MESSAGE_LEN);
         }
     }
 
@@ -2341,7 +2373,7 @@ mod tests {
         let chunks = split_message(&text);
         assert!(chunks.len() >= 2);
         for chunk in &chunks {
-            assert!(chunk.chars().count() <= TELEGRAM_MAX_MESSAGE_LEN);
+            assert!(utf16_len(chunk) <= TELEGRAM_MAX_MESSAGE_LEN);
         }
         // Rejoined must preserve all characters
         let rejoined: String = chunks.concat();
@@ -2358,12 +2390,25 @@ mod tests {
         let chunks = split_message(&text);
         assert!(chunks.len() >= 2);
         for chunk in &chunks {
-            assert!(chunk.chars().count() <= TELEGRAM_MAX_MESSAGE_LEN);
+            assert!(utf16_len(chunk) <= TELEGRAM_MAX_MESSAGE_LEN);
             // Every char should be a complete emoji
             assert!(chunk.chars().all(|c| c == '\u{1F600}'));
         }
     }
 
+    #[test]
+    fn test_split_message_exact_utf16_limit_for_surrogate_pairs() {
+        let emoji = "\u{1F600}"; // 😀
+        let text = emoji.repeat(TELEGRAM_MAX_MESSAGE_LEN);
+
+        let chunks = split_message(&text);
+
+        assert_eq!(chunks.len(), 2);
+        assert!(chunks
+            .iter()
+            .all(|chunk| utf16_len(chunk) <= TELEGRAM_MAX_MESSAGE_LEN));
+    }
+
     #[test]
     fn test_clean_message_text() {
         // Without bot_username: strips any leading @mention

crates/ironclaw_gateway/static/app.js

@@ -7084,6 +7084,12 @@ function loadSettingsSubtab(subtab) {
 // --- Structured Settings Definitions ---
 
 var INFERENCE_SETTINGS = [
+  {
+    group: 'cfg.group.inference',
+    settings: [
+      { key: 'temperature', label: 'cfg.temperature.label', description: 'cfg.temperature.desc', type: 'float', min: 0, max: 2, step: 0.1 },
+    ]
+  },
   {
     group: 'cfg.group.embeddings',
     settings: [
@@ -7429,25 +7435,25 @@ function renderStructuredSettingsRow(def, value, activeValue) {
       return function() { saveSetting(k, el.value === '' ? null : el.value); };
     })(def.key, sel));
     inputWrap.appendChild(sel);
-  } else if (def.type === 'number') {
+  } else if (def.type === 'number' || def.type === 'float') {
     var numInp = document.createElement('input');
     numInp.type = 'number';
-    numInp.step = '1';
+    numInp.step = def.step !== undefined ? String(def.step) : (def.type === 'float' ? 'any' : '1');
     numInp.className = 'settings-input';
     numInp.setAttribute('aria-label', ariaLabel);
     numInp.value = (value === null || value === undefined) ? '' : value;
     if (!value && value !== 0) numInp.placeholder = placeholderText;
     if (def.min !== undefined) numInp.min = def.min;
     if (def.max !== undefined) numInp.max = def.max;
-    numInp.addEventListener('change', (function(k, el) {
+    numInp.addEventListener('change', (function(k, el, isFloat) {
       return function() {
         if (el.value === '') return saveSetting(k, null);
-        var parsed = parseInt(el.value, 10);
+        var parsed = isFloat ? parseFloat(el.value) : parseInt(el.value, 10);
         if (isNaN(parsed)) return;
         el.value = parsed;
         saveSetting(k, parsed);
       };
-    })(def.key, numInp));
+    })(def.key, numInp, def.type === 'float'));
     inputWrap.appendChild(numInp);
   } else if (def.type === 'list') {
     var listInp = document.createElement('input');

crates/ironclaw_gateway/static/i18n/en.js

@@ -521,6 +521,9 @@ I18n.register('en', {
   'cfg.llm_backend.desc': 'LLM inference provider',
   'cfg.selected_model.label': 'Model',
   'cfg.selected_model.desc': 'Model name or ID for the selected backend',
+  'cfg.temperature.label': 'Temperature',
+  'cfg.temperature.desc': 'Default sampling temperature (0.0–2.0). Lower = more deterministic, higher = more creative',
+  'cfg.group.inference': 'Inference',
   'cfg.ollama_base_url.label': 'Ollama URL',
   'cfg.ollama_base_url.desc': 'Base URL for Ollama API',
   'cfg.openai_compatible_base_url.label': 'OpenAI-compatible URL',

crates/ironclaw_gateway/static/i18n/ko.js

@@ -520,6 +520,9 @@ I18n.register('ko', {
   'cfg.llm_backend.desc': 'LLM 추론 공급자',
   'cfg.selected_model.label': '모델',
   'cfg.selected_model.desc': '선택한 백엔드의 모델 이름 또는 ID',
+  'cfg.temperature.label': '온도',
+  'cfg.temperature.desc': '기본 샘플링 온도 (0.0–2.0). 낮을수록 결정적, 높을수록 창의적',
+  'cfg.group.inference': '추론',
   'cfg.ollama_base_url.label': 'Ollama URL',
   'cfg.ollama_base_url.desc': 'Ollama API의 베이스 URL',
   'cfg.openai_compatible_base_url.label': 'OpenAI 호환 URL',

crates/ironclaw_gateway/static/i18n/zh-CN.js

@@ -520,6 +520,9 @@ I18n.register('zh-CN', {
   'cfg.llm_backend.desc': 'LLM 推理提供商',
   'cfg.selected_model.label': '模型',
   'cfg.selected_model.desc': '所选后端的模型名称或 ID',
+  'cfg.temperature.label': '温度',
+  'cfg.temperature.desc': '默认采样温度（0.0–2.0）。越低越确定性，越高越有创意',
+  'cfg.group.inference': '推理',
   'cfg.ollama_base_url.label': 'Ollama URL',
   'cfg.ollama_base_url.desc': 'Ollama API 基础 URL',
   'cfg.openai_compatible_base_url.label': 'OpenAI 兼容 URL',