@gaberrb/polypus 0.4.9 → 0.4.10

package/dist/index.js

@@ -142,6 +142,7 @@ var en = {
   "run.reprompt": "\u21BB no tool call \u2014 reinforcing instructions (attempt {attempt})",
   "run.autocorrect": "\u21BB tool failed \u2014 auto-correcting with extra context",
   "run.cancelled": "\u25A0 cancelled",
+  "compaction.done": "context compacted: ~{before} \u2192 ~{after} tokens",
   "run.jsonNeedsTask": "--json requires a task argument (headless mode has no interactive REPL).",
   "review.approveAll": "approve all",
   "review.reject": "reject",
@@ -401,6 +402,7 @@ var ptBR = {
   "run.reprompt": "\u21BB nenhuma chamada de tool \u2014 refor\xE7ando instru\xE7\xF5es (tentativa {attempt})",
   "run.autocorrect": "\u21BB tool falhou \u2014 autocorrigindo com contexto extra",
   "run.cancelled": "\u25A0 cancelado",
+  "compaction.done": "contexto compactado: ~{before} \u2192 ~{after} tokens",
   "run.jsonNeedsTask": "--json exige um argumento de tarefa (o modo headless n\xE3o tem REPL interativo).",
   "review.approveAll": "aprovar tudo",
   "review.reject": "rejeitar",
@@ -2092,6 +2094,56 @@ async function loadProjectInstructions(workspace) {
   return void 0;
 }
 
+// src/core/agent/compaction.ts
+function estimateTokens(messages) {
+  let chars = 0;
+  for (const m of messages) chars += m.content.length;
+  return Math.ceil(chars / 4);
+}
+var RECENT_KEEP = 8;
+var MIN_TO_COMPACT = 4;
+var MAX_SUMMARY_INPUT = 4e4;
+function findSafeCut(messages, desiredKeep = RECENT_KEEP) {
+  let cut = Math.max(1, messages.length - desiredKeep);
+  while (cut < messages.length && (messages[cut].role === "tool" || messages[cut - 1]?.role === "assistant" && (messages[cut - 1].toolCalls?.length ?? 0) > 0)) {
+    cut++;
+  }
+  return cut;
+}
+function serialize(messages) {
+  const text2 = messages.map((m) => {
+    const tools = m.toolCalls?.length ? ` [called: ${m.toolCalls.map((c) => c.name).join(", ")}]` : "";
+    return `${m.role}${tools}: ${m.content}`;
+  }).join("\n\n");
+  return text2.length > MAX_SUMMARY_INPUT ? text2.slice(-MAX_SUMMARY_INPUT) : text2;
+}
+async function compactHistory(messages, agent, signal) {
+  if (messages.length === 0) return messages;
+  const system = messages[0].role === "system" ? messages[0] : void 0;
+  const startIdx = system ? 1 : 0;
+  const cut = findSafeCut(messages);
+  if (cut >= messages.length) return messages;
+  const middle = messages.slice(startIdx, cut);
+  if (middle.length < MIN_TO_COMPACT) return messages;
+  const tail = messages.slice(cut);
+  const summary = await agent.provider.chat({
+    messages: [
+      {
+        role: "system",
+        content: "You compress a coding agent's conversation so it can continue with less context. Summarize the messages below into a concise but information-dense brief that preserves: the original task and goal, key decisions, files created/edited and why, important command/test outputs, and any remaining TODOs or open problems. Use terse bullet points. Do not invent details."
+      },
+      { role: "user", content: serialize(middle) }
+    ],
+    signal
+  });
+  const summaryMessage = {
+    role: "user",
+    content: `[Summary of earlier conversation, compacted to save context]
+${summary.content.trim()}`
+  };
+  return system ? [system, summaryMessage, ...tail] : [summaryMessage, ...tail];
+}
+
 // src/core/agent/loop.ts
 function looksLikeStall(text2) {
   const lc = text2.toLowerCase();
@@ -2146,9 +2198,22 @@ async function runAgent(opts) {
   const maxToolRetries = opts.maxToolRetries ?? 3;
   const autoCorrect = opts.autoCorrect ?? true;
   const usage2 = { promptTokens: 0, completionTokens: 0 };
+  const compactThreshold = opts.compactThresholdTokens ?? 0;
+  let lastPromptTokens = 0;
   for (let step = 1; step <= maxSteps; step++) {
     if (opts.signal?.aborted) return { finished: false, reason: "cancelled", steps: step - 1, messages, usage: usage2 };
     events?.onStep?.(step);
+    if (compactThreshold > 0) {
+      const current = lastPromptTokens || estimateTokens(messages);
+      if (current >= compactThreshold) {
+        const compacted = await compactHistory(messages, agent, opts.signal);
+        if (compacted.length < messages.length) {
+          messages.splice(0, messages.length, ...compacted);
+          lastPromptTokens = estimateTokens(messages);
+          events?.onCompaction?.(current, lastPromptTokens);
+        }
+      }
+    }
     let response;
     try {
       response = await agent.provider.chat({
@@ -2163,6 +2228,7 @@ async function runAgent(opts) {
     }
     usage2.promptTokens += response.usage?.promptTokens ?? 0;
     usage2.completionTokens += response.usage?.completionTokens ?? 0;
+    lastPromptTokens = response.usage?.promptTokens ?? estimateTokens(messages);
     events?.onUsage?.(usage2);
     const { toolCalls, text: text2 } = driver.parse(response);
     messages.push(driver.assistantMessage(response, toolCalls));
@@ -2588,6 +2654,9 @@ function createJsonCollector() {
     onReprompt(attempt) {
       log.push({ type: "reprompt", attempt });
     },
+    onCompaction(before, after) {
+      log.push({ type: "compaction", before, after });
+    },
     onUsage() {
     }
   };
@@ -3800,6 +3869,11 @@ var Spinner = class {
 
 // src/cli/commands/run.ts
 var MAX_VERIFY_FIXES = 3;
+function compactionThreshold() {
+  if (process.env.POLYPUS_NO_COMPACT) return 0;
+  const v = Number(process.env.POLYPUS_COMPACT_THRESHOLD);
+  return Number.isFinite(v) && v > 0 ? v : 12e4;
+}
 async function run(task, opts) {
   let config = await loadConfig();
   const workspace = process.cwd();
@@ -3925,6 +3999,7 @@ async function executeTask(task, resolved, workspace, session, json = false, ver
     promptContext: { workspace, mode: session.mode, allow: session.allow },
     history: session.history,
     maxSteps: session.maxSteps,
+    compactThresholdTokens: compactionThreshold(),
     signal: controller.signal,
     events
   });
@@ -4108,6 +4183,10 @@ function renderEvents(spinner3) {
       spinner3.stop();
       console.log(pc8.yellow("  " + t("run.reprompt", { attempt })));
     },
+    onCompaction(before, after) {
+      spinner3.stop();
+      console.log(pc8.dim("\u21AF " + t("compaction.done", { before: fmtTokens(before), after: fmtTokens(after) })));
+    },
     onCorrection() {
       spinner3.stop();
       console.log(pc8.yellow("    \u21BB " + t("run.autocorrect")));