kiosapi 0.1.17 → 0.1.18

package/dist/agent/run.js

@@ -89,8 +89,27 @@ export function loadCheckpoint() {
         return null;
     try {
         const data = JSON.parse(readFileSync(CHECKPOINT_PATH, 'utf8'));
+        // Sanitize messages from older checkpoint formats:
+        //  - Strip extra system messages (index > 0): pre-v0.1.17 auto-inject pushed a second
+        //    {role:'system'} that some providers reject when they appear mid-conversation.
+        //  - Null out content when tool_calls are present: strict OpenAI-compat providers (DeepSeek
+        //    etc.) require content:null on assistant tool-call turns; old checkpoints stored the
+        //    prefacing text alongside tool_calls which caused HTTP 400 on the next API call.
+        const [sys0, ...rest] = data.messages ?? [];
+        const sanitized = [
+            ...(sys0 ? [sys0] : []),
+            ...rest
+                .filter((m) => m.role !== 'system')
+                .map((m) => {
+                if (m.role === 'assistant' && m.tool_calls?.length) {
+                    return { ...m, content: null };
+                }
+                return m;
+            }),
+        ];
         return {
             ...data,
+            messages: sanitized,
             teamModels: data.teamModels ?? {},
             totalTokens: data.totalTokens ?? 0,
         };
@@ -373,22 +392,28 @@ export function undoLastTurn(s) {
  * agent's final text (last answer or the `selesai` summary) — useful for chaining agents in a team.
  */
 export async function runTurn(s, userText) {
-    // On the very first turn of a fresh session, append project metadata to the existing system
-    // message so the model starts oriented without needing daftar_file(".") for basic orientation.
-    // We APPEND (not push a new system message) because some providers reject requests that have
-    // more than one system-role message — a second {role:'system'} at index 1 causes HTTP 400.
+    // On the very first turn of a fresh session, append a compact project snapshot to the
+    // existing system message so the model starts oriented without calling daftar_file(".").
+    // DESIGN CONSTRAINTS:
+    //  - APPEND, never push a second system message (some providers reject role:'system' at idx>0)
+    //  - Keep total injected content small: free-tier models (Workers AI llama etc.) have 8K token
+    //    context windows; a fat system message plus a few tool results fills it up fast, causing the
+    //    upstream to return HTTP 400. Depth-2 tree (~1-2KB) is enough for high-level orientation.
     if (s.messages.filter((m) => m.role === 'user').length === 0) {
         const snippets = [];
+        // Depth 2 (not 3): shows top-level dirs + their immediate contents — enough to orient without
+        // producing 300 lines that consume half a small model's context window.
         try {
-            snippets.push(`<root-directory>\n${daftarFile('.')}\n</root-directory>`);
+            snippets.push(`<root-directory>\n${daftarFile('.', 2)}\n</root-directory>`);
         }
         catch { /* ignore */ }
-        for (const f of ['package.json', 'pyproject.toml', 'Cargo.toml', 'go.mod', 'README.md']) {
+        for (const f of ['package.json', 'pyproject.toml', 'Cargo.toml', 'go.mod']) {
             const abs = join(process.cwd(), f);
             if (existsSync(abs)) {
                 try {
                     const content = readFileSync(abs, 'utf8');
-                    snippets.push(`<file path="${f}">\n${content.slice(0, 4000)}\n</file>`);
+                    // 1500 chars: enough for name/scripts/deps, not the full lockfile prose
+                    snippets.push(`<file path="${f}">\n${content.slice(0, 1_500)}\n</file>`);
                 }
                 catch { /* unreadable — skip */ }
             }
@@ -396,8 +421,11 @@ export async function runTurn(s, userText) {
         if (snippets.length > 0) {
             const sysMsg = s.messages[0];
             if (sysMsg?.role === 'system') {
+                // Hard cap: keep total injected context under 5KB so the system message stays manageable.
+                const joined = snippets.join('\n\n');
+                const capped = joined.length > 5_000 ? `${joined.slice(0, 5_000)}\n…` : joined;
                 sysMsg.content +=
-                    `\n\n## Konteks Proyek (auto-injected — jangan panggil daftar_file(".") lagi)\n${snippets.join('\n\n')}`;
+                    `\n\n## Konteks Proyek\n${capped}`;
             }
         }
     }
@@ -527,11 +555,19 @@ export async function runTurn(s, userText) {
                 continue;
             }
             const result = await runTool(call, s.otomatis, s.model);
-            s.messages.push({ role: 'tool', content: result.output, tool_call_id: call.id });
+            // Cap what goes into the conversation history: full output can be 300 lines of directory
+            // listing or 100 KB of file content, which quickly overflows small-context-window models
+            // (Workers AI llama has ~8K token limit). The full result is already returned by runTool;
+            // truncating only the *stored* copy keeps the API payload manageable without losing info.
+            const MAX_STORED_RESULT = 5_000;
+            const stored = result.output.length > MAX_STORED_RESULT
+                ? `${result.output.slice(0, MAX_STORED_RESULT)}\n…[dipotong — gunakan path/range spesifik jika perlu lebih]`
+                : result.output;
+            s.messages.push({ role: 'tool', content: stored, tool_call_id: call.id });
             if (result.modifiedPath)
                 stepModified.add(result.modifiedPath);
             if (READ_ONLY_TOOLS.has(call.function.name))
-                toolCache.set(sig, result.output);
+                toolCache.set(sig, stored);
             if (result.done) {
                 if (stepModified.size > 0)
                     console.log(dim(`  ✎ ${[...stepModified].join(' · ')}`));

package/dist/agent/schemas.js

@@ -198,7 +198,8 @@ Direktori kerja: ${process.cwd()}
 OS: ${osName} · ${shellNote}
 Aturan:
 - Bekerja langkah demi langkah: pakai tool untuk membaca sebelum mengubah.
-- Strategi eksplorasi: daftar_file(".") sudah tersedia di konteks awal — identifikasi file yang relevan LANGSUNG, lalu baca dengan baca_file. Jangan ulangi daftar_file pada path yang sama.
+- Strategi eksplorasi: struktur root tersedia di konteks awal — identifikasi subfolder relevan LANGSUNG, lalu baca dengan baca_file. Jangan ulangi daftar_file pada path yang sama.
+- Kedalaman daftar_file: gunakan kedalaman=2 (default) untuk subfolder besar. Kedalaman=3 hanya jika kamu sudah tahu folder itu kecil. JANGAN gunakan kedalaman=4+ kecuali diminta eksplisit.
 - JANGAN memanggil tool APAPUN dengan argumen identik lebih dari 1× dalam satu sesi. Jika tool mengembalikan peringatan cache "⚠", langsung ganti ke path atau argumen BERBEDA.
 - Path selalu relatif ke direktori kerja; akses ke luar ditolak.
 - Gunakan hapus_file/pindah_file untuk menghapus/memindahkan file (lebih aman dari jalankan del/rm).

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "kiosapi",
-  "version": "0.1.17",
+  "version": "0.1.18",
   "type": "module",
   "description": "CLI Kiosapi.id berbahasa Indonesia — bangun aplikasimu pakai API key Kiosapi (agen + multimodal).",
   "keywords": [