zelari-code 2.18.1 → 2.20.0

package/dist/cli/main.bundled.js

@@ -29977,8 +29977,9 @@ var init_service = __esm({
         const text = (raw.query ?? raw.text ?? "").slice(0, 64e3);
         const limit = Math.max(1, Math.min(raw.limit ?? 8, 100));
         this.emit({ type: "memory_recall_start" });
+        const { weights, ...recallQuery } = raw;
         const candidates = await this.backend.search({
-          ...raw,
+          ...recallQuery,
           text,
           projectId: this.projectId,
           statuses: raw.statuses ?? (raw.includeHistorical ? void 0 : ["active"]),
@@ -30025,7 +30026,7 @@ var init_service = __esm({
             }
           }
         }
-        const ranked = diverse(rankMemoryCandidates([...byId.values()], text), limit);
+        const ranked = diverse(rankMemoryCandidates([...byId.values()], text, weights ? { weights } : {}), limit);
         this.emit({
           type: "memory_recall_end",
           durationMs: Date.now() - started,
@@ -40721,25 +40722,883 @@ var init_skillsMd = __esm({
   }
 });
 
+// src/cli/budget/historySummary.ts
+function extractiveHistorySummary(dropped, opts) {
+  const maxChars = opts?.maxChars ?? MAX_SUMMARY_CHARS;
+  if (dropped.length === 0) return "No prior turns.";
+  const userGoals = [];
+  const assistantNotes = [];
+  const userConstraints = [];
+  const unresolved = [];
+  const verification = [];
+  const decisions = [];
+  const tools = /* @__PURE__ */ new Map();
+  const files = /* @__PURE__ */ new Set();
+  let toolResults = 0;
+  for (const m of dropped) {
+    if (m.role === "user" && m.content.trim()) {
+      const goal = oneLine(m.content, 220);
+      userGoals.push(goal);
+      if (/\b(must|never|required|only|do not|constraint|vincolo|deve|senza)\b/i.test(goal)) userConstraints.push(goal);
+    } else if (m.role === "assistant") {
+      if (m.content.trim()) {
+        const note = oneLine(m.content, 220);
+        assistantNotes.push(note);
+        if (/\b(fail|failed|error|unresolved|remaining|todo|blocked|gap|errore|fallit|irrisolt|manca)\b/i.test(note)) unresolved.push(note);
+        if (/\b(test|typecheck|build|verify|verification|passed|failed|green|red)\b/i.test(note)) verification.push(note);
+        if (/\b(decid|decision|chosen|choose|scelt|adopt|implement)\b/i.test(note)) decisions.push(note);
+      }
+      if (m.toolCalls) {
+        for (const tc of m.toolCalls) {
+          tools.set(tc.name, (tools.get(tc.name) ?? 0) + 1);
+          collectPaths3(tc.args, files);
+        }
+      }
+    } else if (m.role === "tool") {
+      toolResults += 1;
+      collectPathsFromText(m.content, files);
+      if (/\b(fail|failed|error|exception|blocked|errore|fallit)\b/i.test(m.content)) unresolved.push(oneLine(m.content, 220));
+      if (/\b(test|typecheck|build|verify|passed|failed|success)\b/i.test(m.content)) verification.push(oneLine(m.content, 220));
+    }
+  }
+  const parts = [
+    "[history-summary] Earlier turns were compacted to stay within the context budget.",
+    `Dropped ${dropped.length} message(s) (${userGoals.length} user, ${assistantNotes.length} assistant notes, ${toolResults} tool results).`
+  ];
+  if (userGoals.length) {
+    parts.push("## User goals / requests");
+    for (const g of userGoals.slice(-6)) parts.push(`- ${g}`);
+  }
+  if (userConstraints.length) {
+    parts.push("## User constraints (preserve exactly)");
+    for (const item of [...new Set(userConstraints)].slice(-8)) parts.push(`- ${item}`);
+  }
+  if (unresolved.length) {
+    parts.push("## Unresolved failures / pending repair");
+    for (const item of [...new Set(unresolved)].slice(-8)) parts.push(`- ${item}`);
+  }
+  if (verification.length) {
+    parts.push("## Latest verification state");
+    for (const item of [...new Set(verification)].slice(-6)) parts.push(`- ${item}`);
+  }
+  if (decisions.length) {
+    parts.push("## Recent active decisions");
+    for (const item of [...new Set(decisions)].slice(-6)) parts.push(`- ${item}`);
+  }
+  if (assistantNotes.length) {
+    parts.push("## Assistant conclusions (truncated)");
+    for (const a of assistantNotes.slice(-5)) parts.push(`- ${a}`);
+  }
+  if (tools.size) {
+    const ranked = [...tools.entries()].sort((a, b) => b[1] - a[1]).slice(0, 12).map(([n, c]) => `${n}\xD7${c}`);
+    parts.push(`## Tools used: ${ranked.join(", ")}`);
+  }
+  if (files.size) {
+    const list = [...files].slice(0, 24);
+    parts.push(`## Paths mentioned: ${list.join(", ")}`);
+  }
+  parts.push(
+    "Continue from the recent messages below; do not re-ask goals already answered above unless the user changes them."
+  );
+  let out = parts.join("\n");
+  if (out.length > maxChars) {
+    out = `${out.slice(0, maxChars - 1)}\u2026`;
+  }
+  return out;
+}
+function oneLine(s, max) {
+  const t = s.replace(/\s+/g, " ").trim();
+  if (t.length <= max) return t;
+  return `${t.slice(0, max - 1)}\u2026`;
+}
+function collectPaths3(args, out) {
+  if (!args || typeof args !== "object") return;
+  const obj = args;
+  for (const key of ["path", "file", "filepath", "filePath", "target", "cwd"]) {
+    const v = obj[key];
+    if (typeof v === "string" && v.length > 1 && v.length < 260) {
+      out.add(v.replace(/\\/g, "/"));
+    }
+  }
+  if (typeof obj.file_path === "string") out.add(String(obj.file_path));
+}
+function collectPathsFromText(text, out) {
+  const re = /(?:^|[\s"'`])((?:[\w.-]+\/)+[\w.-]+\.\w{1,8})/g;
+  let m;
+  let n = 0;
+  while ((m = re.exec(text)) !== null && n < 8) {
+    out.add(m[1]);
+    n += 1;
+  }
+}
+var MAX_SUMMARY_CHARS;
+var init_historySummary = __esm({
+  "src/cli/budget/historySummary.ts"() {
+    "use strict";
+    MAX_SUMMARY_CHARS = 3500;
+  }
+});
+
+// src/cli/budget/llmCompact.ts
+function isLlmCompactEnabled() {
+  const v = process.env.ZELARI_LLM_COMPACT?.trim().toLowerCase();
+  if (v === "0" || v === "false" || v === "off" || v === "no") return false;
+  return true;
+}
+function compactModelOverride() {
+  const v = process.env.ZELARI_COMPACT_MODEL?.trim();
+  return v ? v : void 0;
+}
+async function llmSummarizeHistoryReplay(input) {
+  const override = input.overrideModel ?? compactModelOverride();
+  const model = override ?? input.model;
+  const cacheReuseExpected = !override;
+  if (!isLlmCompactEnabled()) return { summary: null, model, cacheReuseExpected };
+  if (input.droppedMessages.length === 0) {
+    return { summary: null, model, cacheReuseExpected };
+  }
+  const messages = [
+    ...input.systemMessages,
+    ...input.droppedMessages,
+    {
+      role: "user",
+      content: COMPACTION_INSTRUCTION
+    }
+  ];
+  const controller = new AbortController();
+  const timeout = setTimeout(() => controller.abort(), REPLAY_TIMEOUT_MS);
+  const onOuterAbort = () => controller.abort();
+  input.signal?.addEventListener("abort", onOuterAbort, { once: true });
+  try {
+    let text = "";
+    let emittedToolCall = false;
+    for await (const delta of input.providerStream({
+      provider: input.provider,
+      model,
+      messages,
+      // Tools stay advertised: dropping them would change the prefix token
+      // sequence and destroy cache reuse (explicit DSH decision). They are
+      // sorted canonically (same discipline as the live routed request and
+      // the snapshot fingerprints) so the replay prefix is byte-identical.
+      tools: [...input.tools].sort((a, b) => a.name < b.name ? -1 : a.name > b.name ? 1 : 0),
+      signal: controller.signal,
+      generation: {
+        purpose: "compaction",
+        temperature: 0.1,
+        maxTokens: 900
+      }
+    })) {
+      if (delta.kind === "text") text += delta.delta;
+      if (delta.kind === "tool_call") emittedToolCall = true;
+    }
+    if (emittedToolCall) return { summary: null, model, cacheReuseExpected };
+    if (!text.trim()) return { summary: null, model, cacheReuseExpected };
+    return { summary: text.trim(), model, cacheReuseExpected };
+  } catch {
+    return { summary: null, model, cacheReuseExpected };
+  } finally {
+    clearTimeout(timeout);
+    input.signal?.removeEventListener("abort", onOuterAbort);
+  }
+}
+var COMPACTION_INSTRUCTION, REPLAY_TIMEOUT_MS;
+var init_llmCompact = __esm({
+  "src/cli/budget/llmCompact.ts"() {
+    "use strict";
+    COMPACTION_INSTRUCTION = `
+You are now acting as a compaction engine for this coding-agent session.
+
+Condense the conversation ABOVE into a compact checkpoint sufficient to continue the task.
+
+Preserve:
+- user's goal and evolving intent
+- decisions already made
+- exact file paths and identifiers
+- code changes already completed
+- commands/errors that still matter
+- constraints
+- unfinished work
+- the single most likely next action
+
+Do not call tools.
+Do not mention this summarization request.
+Output only the checkpoint.
+Be concise.
+`.trim();
+    REPLAY_TIMEOUT_MS = 6e4;
+  }
+});
+
+// src/cli/hooks/historyCompaction.ts
+function compactedRangeFromDropped(dropped) {
+  if (dropped.length === 0) return void 0;
+  const seqs = [];
+  const sources = [];
+  for (const m of dropped) {
+    const hasCompactRange = typeof m.compactedFromSeq === "number" && Number.isInteger(m.compactedFromSeq) && m.compactedFromSeq > 0 && typeof m.compactedToSeq === "number" && Number.isInteger(m.compactedToSeq) && m.compactedToSeq >= m.compactedFromSeq;
+    if (hasCompactRange) {
+      seqs.push(m.compactedFromSeq, m.compactedToSeq);
+      sources.push(...m.sourceEventSeqs ?? []);
+      if (typeof m.seq === "number" && Number.isInteger(m.seq) && m.seq > 0) {
+        seqs.push(m.seq);
+        sources.push(m.seq);
+      }
+      continue;
+    }
+    if (typeof m.seq !== "number" || !Number.isInteger(m.seq) || m.seq < 1) return void 0;
+    seqs.push(m.seq);
+    sources.push(m.seq);
+  }
+  return {
+    fromSeq: Math.min(...seqs),
+    toSeq: Math.max(...seqs),
+    sourceEventSeqs: [...new Set(sources)]
+  };
+}
+function withDroppedRange(result, dropped, strategy) {
+  const range = compactedRangeFromDropped(dropped);
+  if (!range) return { ...result, strategy };
+  return { ...result, ...range, strategy };
+}
+function resolveMaxMessages(opts) {
+  const envTurns = envNumber(process.env.ZELARI_HISTORY_TURNS, { default: 6, min: 0 });
+  let turns = opts?.maxMessages ? Math.ceil(opts.maxMessages / 4) : envTurns;
+  if (opts?.durableStatePresent && !opts?.maxMessages && process.env.ZELARI_HISTORY_TURNS === void 0) {
+    turns = Math.min(turns, 3);
+  }
+  if (turns <= 0) return 0;
+  if (opts?.force && opts?.maxMessages) return Math.max(1, opts.maxMessages);
+  return turns * 4;
+}
+function findValidCutIndex(messages, naiveCut) {
+  let cut = naiveCut;
+  while (cut < messages.length) {
+    const kept = messages.slice(cut);
+    const declared = /* @__PURE__ */ new Set();
+    for (const m of kept) {
+      if (m.role === "assistant" && m.toolCalls) {
+        for (const tc of m.toolCalls) declared.add(tc.id);
+      }
+    }
+    let moved = false;
+    for (let k = 0; k < kept.length; k++) {
+      const m = kept[k];
+      if (m.role === "tool" && m.toolCallId && !declared.has(m.toolCallId)) {
+        for (let j = cut - 1; j >= 0; j--) {
+          const prev2 = messages[j];
+          if (prev2.role === "assistant" && prev2.toolCalls && prev2.toolCalls.some((tc) => tc.id === m.toolCallId)) {
+            cut = j;
+            moved = true;
+            break;
+          }
+        }
+        break;
+      }
+    }
+    if (!moved) break;
+  }
+  return cut;
+}
+function resolvePruneLimits(opts) {
+  const maxChars = opts?.maxChars ?? envNumber(process.env.ZELARI_TOOL_RESULT_MAX_CHARS, { default: 8e3, min: 256 });
+  const rawTail = opts?.tailChars ?? envNumber(process.env.ZELARI_TOOL_RESULT_TAIL_CHARS, { default: 1e3, min: 0 });
+  const tailChars = Math.min(rawTail, maxChars);
+  return { maxChars, tailChars };
+}
+function pruneToolResultsDetailed(messages, opts) {
+  const { maxChars, tailChars } = resolvePruneLimits(opts);
+  const headChars = maxChars - tailChars;
+  const stats = { pruned: 0, charsOmitted: 0 };
+  let changed = false;
+  const out = messages.map((m) => {
+    if (m.role !== "tool") return m;
+    const body = m.content ?? "";
+    if (body.length <= maxChars) return m;
+    const head = headChars > 0 ? body.slice(0, headChars) : "";
+    const tail2 = tailChars > 0 ? body.slice(-tailChars) : "";
+    const omitted = body.length - head.length - tail2.length;
+    changed = true;
+    stats.pruned += 1;
+    stats.charsOmitted += omitted;
+    return {
+      ...m,
+      content: [head, "\u2026[pruned " + omitted + " chars]\u2026", tail2].join(String.fromCharCode(10))
+    };
+  });
+  return {
+    messages: changed ? out : messages,
+    stats
+  };
+}
+function compactHistory(messages, opts) {
+  return compactHistoryDetailed(messages, opts).messages;
+}
+function buildCheckpointMessage(summaryText, range) {
+  return {
+    role: "user",
+    content: CHECKPOINT_WRAPPER_PREFIX + "\n\n<compacted-summary>\n" + summaryText + "\n</compacted-summary>",
+    ...range ? {
+      compactedFromSeq: range.fromSeq,
+      compactedToSeq: range.toSeq,
+      sourceEventSeqs: [...range.sourceEventSeqs]
+    } : {}
+  };
+}
+function compactHistoryDetailed(messages, opts) {
+  const maxMessages = resolveMaxMessages(opts);
+  if (maxMessages === 0) {
+    return withDroppedRange(
+      { messages: [], compacted: true, messagesRemoved: messages.length, summary: "" },
+      messages,
+      "extractive"
+    );
+  }
+  if (messages.length <= maxMessages * 2 && !opts?.force) {
+    return {
+      messages,
+      compacted: false,
+      messagesRemoved: 0,
+      summary: ""
+    };
+  }
+  const naiveCut = Math.max(0, messages.length - maxMessages);
+  const cut = findValidCutIndex(messages, naiveCut);
+  if (cut === 0) {
+    return {
+      messages,
+      compacted: false,
+      messagesRemoved: 0,
+      summary: ""
+    };
+  }
+  const droppedMsgs = messages.slice(0, cut);
+  const droppedRange = compactedRangeFromDropped(droppedMsgs);
+  const pruned = pruneToolResultsDetailed(messages.slice(cut));
+  const kept = pruned.messages;
+  const summaryText = extractiveHistorySummary(droppedMsgs);
+  const summary = buildCheckpointMessage(
+    summaryText || `${COMPACT_MARKER} ${cut} earlier message(s) dropped.`,
+    droppedRange
+  );
+  return withDroppedRange(
+    {
+      messages: [summary, ...kept],
+      compacted: true,
+      messagesRemoved: cut,
+      summary: summary.content,
+      prunedToolResults: pruned.stats.pruned
+    },
+    droppedMsgs,
+    "extractive"
+  );
+}
+async function compactHistoryAsync(messages, opts) {
+  const base2 = compactHistoryDetailed(messages, opts);
+  if (!base2.compacted || base2.messagesRemoved === 0) return base2;
+  const cut = base2.messagesRemoved;
+  const droppedMsgs = messages.slice(0, cut);
+  const extractive = extractiveHistorySummary(droppedMsgs);
+  let summaryText = extractive;
+  let cacheReuseExpected;
+  let replayExactPrefix;
+  const canReplay = !!(opts?.providerStream && opts?.requestSnapshot);
+  if (canReplay) {
+    try {
+      const replay = await llmSummarizeHistoryReplay({
+        providerStream: opts.providerStream,
+        provider: opts.requestSnapshot.provider,
+        model: opts.requestSnapshot.model,
+        systemMessages: opts.requestSnapshot.systemMessages,
+        tools: opts.requestSnapshot.tools,
+        droppedMessages: droppedMsgs,
+        signal: opts?.signal
+      });
+      cacheReuseExpected = replay.cacheReuseExpected;
+      if (replay.summary && replay.summary.trim().length > 40) {
+        const sourceTokens = roughTokens(droppedMsgs);
+        const summaryTok = Math.ceil(replay.summary.length / 4);
+        if (summaryTok < sourceTokens) {
+          summaryText = replay.summary.trim();
+        }
+      }
+    } catch {
+    }
+  }
+  const pruned = pruneToolResultsDetailed(messages.slice(cut));
+  const kept = pruned.messages;
+  const summary = buildCheckpointMessage(summaryText, compactedRangeFromDropped(droppedMsgs));
+  const usedLlm = summaryText !== extractive && summaryText.trim().length > 40;
+  return withDroppedRange(
+    {
+      messages: [summary, ...kept],
+      compacted: true,
+      messagesRemoved: cut,
+      summary: summaryText,
+      prunedToolResults: pruned.stats.pruned,
+      cacheReuseExpected,
+      replayExactPrefix
+    },
+    droppedMsgs,
+    usedLlm ? "llm" : "extractive"
+  );
+}
+function roughTokens(msgs) {
+  let n = 0;
+  for (const m of msgs) {
+    n += Math.ceil((m.content ?? "").length / 4);
+    if (m.toolCalls) {
+      for (const tc of m.toolCalls) {
+        n += Math.ceil(JSON.stringify(tc.args ?? {}).length / 4);
+      }
+    }
+  }
+  return Math.max(1, n);
+}
+var COMPACT_MARKER, CHECKPOINT_WRAPPER_PREFIX;
+var init_historyCompaction = __esm({
+  "src/cli/hooks/historyCompaction.ts"() {
+    "use strict";
+    init_historySummary();
+    init_llmCompact();
+    init_envNumber();
+    COMPACT_MARKER = "[history] Earlier turns were compacted to stay within the context budget.";
+    CHECKPOINT_WRAPPER_PREFIX = "This is an automatically generated checkpoint of earlier conversation. Treat it as established context and continue directly.";
+  }
+});
+
+// src/cli/provider/capabilities.ts
+function frozenProfile(input) {
+  if (input.reasoning.levels) Object.freeze(input.reasoning.levels);
+  Object.freeze(input.reasoning);
+  Object.freeze(input.promptCache);
+  Object.freeze(input.toolCalling);
+  Object.freeze(input.buildRecovery);
+  Object.freeze(input.sampling);
+  Object.freeze(input.compaction);
+  return Object.freeze(input);
+}
+function resolveHarnessProfile(model, providerId) {
+  const provider = providerId?.trim().toLowerCase();
+  if (provider === "deepseek") return "deepseek-v4";
+  if (provider === "grok") return "grok";
+  if (provider === "minimax") return "minimax";
+  if (provider === "glm") return "glm";
+  if (model && DEEPSEEK_RE.test(model)) return "deepseek-v4";
+  if (model && GROK_RE.test(model)) return "grok";
+  if (model && MINIMAX_RE.test(model)) return "minimax";
+  if (model && GLM_RE.test(model)) return "glm";
+  return "default";
+}
+function capabilitiesFor(model, providerId) {
+  switch (resolveHarnessProfile(model, providerId)) {
+    case "deepseek-v4":
+      return DEEPSEEK_V4_CAPS;
+    case "grok":
+      return GROK_CAPS;
+    case "minimax":
+      return model && MINIMAX_M3_RE.test(model) ? MINIMAX_M3_CAPS : MINIMAX_M2_CAPS;
+    case "glm":
+      return GLM_CAPS;
+    default:
+      return DEFAULT_CAPS;
+  }
+}
+var SHARED_COMPACTION, DEFAULT_CAPS, DEEPSEEK_V4_CAPS, GROK_CAPS, MINIMAX_M3_CAPS, MINIMAX_M2_CAPS, GLM_CAPS, DEEPSEEK_RE, GROK_RE, MINIMAX_RE, MINIMAX_M3_RE, GLM_RE;
+var init_capabilities = __esm({
+  "src/cli/provider/capabilities.ts"() {
+    "use strict";
+    SHARED_COMPACTION = { warnAt: 0.7, compactAt: 0.85, hardAt: 0.95 };
+    DEFAULT_CAPS = frozenProfile({
+      contextWindow: 4e5,
+      reasoning: { supported: true, levels: ["low", "medium", "high"], replayReasoning: true },
+      promptCache: { supported: true, pricedCacheRead: false },
+      toolCalling: { parallel: true },
+      buildRecovery: { forceToolChoice: false, maxForcedTurns: 0 },
+      sampling: { temperature: 0.7 },
+      compaction: { ...SHARED_COMPACTION },
+      profile: "default"
+    });
+    DEEPSEEK_V4_CAPS = frozenProfile({
+      contextWindow: 1e6,
+      reasoning: { supported: true, levels: ["high", "max"], replayReasoning: true },
+      promptCache: { supported: true, pricedCacheRead: true },
+      toolCalling: { parallel: true },
+      buildRecovery: { forceToolChoice: false, maxForcedTurns: 0 },
+      sampling: { temperature: 0.7 },
+      compaction: { ...SHARED_COMPACTION },
+      profile: "deepseek-v4"
+    });
+    GROK_CAPS = frozenProfile({
+      contextWindow: 5e5,
+      // grok-4.x truncation fix: explicit conversation ceiling (see interface
+      // doc). Override at runtime via ZELARI_MAX_OUTPUT_TOKENS.
+      maxOutputTokens: 32768,
+      reasoning: {
+        supported: true,
+        levels: ["low", "medium", "high", "xhigh"],
+        replayReasoning: false
+      },
+      promptCache: {
+        supported: true,
+        pricedCacheRead: true,
+        conversationAffinityHeader: "x-grok-conv-id"
+      },
+      toolCalling: { parallel: true },
+      buildRecovery: { forceToolChoice: true, maxForcedTurns: 1 },
+      sampling: { temperature: 0.7 },
+      compaction: { ...SHARED_COMPACTION },
+      profile: "grok"
+    });
+    MINIMAX_M3_CAPS = frozenProfile({
+      contextWindow: 1e6,
+      reasoning: { supported: true, replayReasoning: true },
+      promptCache: { supported: false, pricedCacheRead: false },
+      toolCalling: { parallel: true },
+      buildRecovery: { forceToolChoice: false, maxForcedTurns: 0 },
+      sampling: { temperature: 0.7 },
+      compaction: { ...SHARED_COMPACTION },
+      profile: "minimax"
+    });
+    MINIMAX_M2_CAPS = frozenProfile({
+      contextWindow: 204800,
+      reasoning: { supported: true, replayReasoning: true },
+      promptCache: { supported: false, pricedCacheRead: false },
+      toolCalling: { parallel: true },
+      buildRecovery: { forceToolChoice: false, maxForcedTurns: 0 },
+      sampling: { temperature: 0.7 },
+      compaction: { ...SHARED_COMPACTION },
+      profile: "minimax"
+    });
+    GLM_CAPS = frozenProfile({
+      contextWindow: 2e5,
+      reasoning: { supported: true, replayReasoning: true },
+      promptCache: { supported: true, pricedCacheRead: false },
+      toolCalling: { parallel: true },
+      buildRecovery: { forceToolChoice: false, maxForcedTurns: 0 },
+      sampling: { temperature: 0.7 },
+      compaction: { ...SHARED_COMPACTION },
+      profile: "glm"
+    });
+    DEEPSEEK_RE = /^deepseek-(?:v4(?:\.|-|$)|chat$|reasoner$)/i;
+    GROK_RE = /^grok(?:\.|-|$)/i;
+    MINIMAX_RE = /^minimax(?:\.|-|$)/i;
+    MINIMAX_M3_RE = /^minimax-m3(?:\.|-|$)/i;
+    GLM_RE = /^glm(?:\.|-|$)/i;
+  }
+});
+
+// src/cli/budget/tokenBudget.ts
+function mergeCompactRange(into, r) {
+  if (r.fromSeq !== void 0 && r.toSeq !== void 0) {
+    into.fromSeq = into.fromSeq === void 0 ? r.fromSeq : Math.min(into.fromSeq, r.fromSeq);
+    into.toSeq = into.toSeq === void 0 ? r.toSeq : Math.max(into.toSeq, r.toSeq);
+    if (r.sourceEventSeqs) into.sourceSeqs.push(...r.sourceEventSeqs);
+  }
+  if (r.strategy === "llm") into.strategy = "llm";
+  else if (r.strategy && !into.strategy) into.strategy = r.strategy;
+}
+function estimateTokens2(text) {
+  if (!text) return 0;
+  return Math.max(1, Math.ceil(text.length / 4));
+}
+function estimateHistoryTokens(messages) {
+  let n = 0;
+  for (const m of messages) {
+    n += estimateTokens2(m.content);
+    if (m.toolCalls) {
+      for (const tc of m.toolCalls) {
+        n += estimateTokens2(tc.name) + estimateTokens2(JSON.stringify(tc.args ?? {}));
+      }
+    }
+  }
+  return n;
+}
+function defaultContextLimitForModel(model, provider) {
+  return capabilitiesFor(model, provider).contextWindow;
+}
+function resolveContextLimit(model, provider) {
+  return envNumber(process.env.ZELARI_CONTEXT_LIMIT, {
+    default: defaultContextLimitForModel(model, provider),
+    min: 4e3,
+    max: 2e6
+  });
+}
+function phaseKnobs(phase2) {
+  return {
+    historyTurns: phase2 === "plan" ? envNumber(process.env.ZELARI_HISTORY_TURNS, { default: 8, min: 0 }) : envNumber(process.env.ZELARI_HISTORY_TURNS, { default: 6, min: 0 }),
+    maxToolLoopIterations: phase2 === "plan" ? envNumber(process.env.ZELARI_MAX_TOOL_LOOP_ITERATIONS, { default: 60, min: 1 }) : envNumber(process.env.ZELARI_MAX_TOOL_LOOP_ITERATIONS, {
+      default: 120,
+      min: 1
+    })
+  };
+}
+async function applyBudgetPolicyAsync(history2, phase2, opts) {
+  const contextLimit = resolveContextLimit(opts?.model, opts?.provider);
+  const compact3 = capabilitiesFor(opts?.model, opts?.provider).compaction;
+  const sessionExtra = opts?.sessionTokens ?? 0;
+  const warnings = [];
+  let { historyTurns, maxToolLoopIterations } = phaseKnobs(phase2);
+  const envelope = opts?.requestSnapshot ?? null;
+  const surface = envelope?.snapshot ?? opts?.requestSurface ?? null;
+  const replayBase = surface ? {
+    provider: surface.provider,
+    model: surface.model,
+    systemMessages: surface.systemMessages,
+    tools: surface.tools
+  } : opts?.providerStream ? { provider: "local", model: opts?.model ?? "unknown", systemMessages: [], tools: [] } : null;
+  const headerTokens = surface ? estimateSystemTokensLite(surface.systemMessages) + estimateToolSchemaTokensLite(surface.tools) : 0;
+  const convTokensOf = (h) => estimateConversationTokensLite(h);
+  let hist = history2;
+  let estimated = surface ? headerTokens + convTokensOf(hist) + RESERVED_OUTPUT_TOKENS : estimateHistoryTokens(hist) + sessionExtra;
+  let occupancy = Math.min(1, estimated / contextLimit);
+  let compactSummary = "";
+  let messagesRemoved = 0;
+  let cacheReuseExpected;
+  let prunedTotal = 0;
+  const compactRange = { sourceSeqs: [] };
+  if (occupancy >= compact3.warnAt && occupancy < compact3.compactAt) {
+    warnings.push(
+      `[budget] context ~${Math.round(occupancy * 100)}% full (${estimated}/${contextLimit} tok full-request est.) \u2014 consider /compact or shorter replies.`
+    );
+  }
+  if (occupancy >= 0.8) {
+    const pruned = pruneToolResultsDetailed(hist);
+    if (pruned.stats.pruned > 0) {
+      hist = pruned.messages;
+      prunedTotal += pruned.stats.pruned;
+      estimated = surface ? headerTokens + convTokensOf(hist) + RESERVED_OUTPUT_TOKENS : estimateHistoryTokens(hist) + sessionExtra;
+      occupancy = Math.min(1, estimated / contextLimit);
+      warnings.push(
+        `[budget] pruned ${pruned.stats.pruned} oversized tool result(s) \u2192 ${Math.round(occupancy * 100)}% (${estimated} tok).`
+      );
+    }
+  }
+  const fold = (r, label, forcedTurns) => {
+    hist = applySessionSurface(r.messages);
+    if (r.compacted) {
+      messagesRemoved += r.messagesRemoved;
+      if (r.summary) compactSummary = r.summary;
+      mergeCompactRange(compactRange, r);
+      if (r.cacheReuseExpected !== void 0) {
+        cacheReuseExpected = r.cacheReuseExpected;
+      }
+    }
+    estimated = surface ? headerTokens + convTokensOf(hist) + RESERVED_OUTPUT_TOKENS : estimateHistoryTokens(hist) + sessionExtra;
+    occupancy = Math.min(1, estimated / contextLimit);
+    warnings.push(
+      `[budget] ${label} \u2014 kept ~${forcedTurns} turns (${estimated} tok est.` + (r.messagesRemoved ? `, removed ${r.messagesRemoved} msgs` : "") + (r.cacheReuseExpected === false ? ", cache reuse NOT expected (model override)" : "") + ")."
+    );
+  };
+  if (occupancy >= compact3.compactAt) {
+    const forcedTurns = Math.max(1, Math.floor(historyTurns / 2));
+    historyTurns = forcedTurns;
+    maxToolLoopIterations = Math.min(
+      maxToolLoopIterations,
+      phase2 === "plan" ? 24 : 40
+    );
+    let r = await compactHistoryAsync(hist, {
+      maxMessages: Math.max(2, forcedTurns * 4),
+      force: true,
+      signal: opts?.signal,
+      ...replayBase ? { requestSnapshot: replayBase, providerStream: opts?.providerStream } : {}
+    });
+    if (!r.compacted) {
+      r = await compactHistoryAsync(hist, {
+        maxMessages: 2,
+        force: true,
+        signal: opts?.signal,
+        ...replayBase ? { requestSnapshot: replayBase, providerStream: opts?.providerStream } : {}
+      });
+    }
+    const label = r.cacheReuseExpected === false ? "llm-compact (model override)" : "auto-compact";
+    fold(r, label, forcedTurns);
+  }
+  if (occupancy >= compact3.hardAt) {
+    const hard = await compactHistoryAsync(hist, {
+      maxMessages: 2,
+      force: true,
+      signal: opts?.signal
+    });
+    fold(hard, hard.summary.includes("\xB7 llm") ? "llm-compact" : "HARD trim", 2);
+    historyTurns = 2;
+    maxToolLoopIterations = Math.min(maxToolLoopIterations, 16);
+    warnings.push(
+      "[budget] HARD context pressure (\u226595%) \u2014 prefer /clear or a new session if quality drops."
+    );
+  }
+  const cacheMetricsLine = envelope ? [
+    "compaction meter:",
+    `provider/model: ${envelope.snapshot.provider}/${envelope.snapshot.model}`,
+    `headerFingerprint: ${envelope.snapshot.headerFingerprint.slice(0, 12)}`,
+    `occupancy: ${Math.round(occupancy * 100)}% (${estimated}/${contextLimit})`,
+    ...envelope.usage?.cachedPromptTokens !== void 0 ? [`cachedPromptTokens: ${envelope.usage.cachedPromptTokens}`] : [],
+    ...cacheReuseExpected !== void 0 ? [`cacheReuseExpected: ${cacheReuseExpected}`] : []
+  ].join(" | ") : void 0;
+  return {
+    history: hist,
+    warnings,
+    maxToolLoopIterations,
+    historyTurns,
+    estimatedHistoryTokens: surface ? convTokensOf(hist) : estimated,
+    contextLimit,
+    occupancy,
+    compactSummary: compactSummary || void 0,
+    messagesRemoved: messagesRemoved || void 0,
+    ...compactRange.fromSeq !== void 0 && compactRange.toSeq !== void 0 ? {
+      compactedFromSeq: compactRange.fromSeq,
+      compactedToSeq: compactRange.toSeq,
+      compactSourceSeqs: compactRange.sourceSeqs,
+      compactStrategy: compactRange.strategy
+    } : {},
+    ...surface ? { contextPressureTokens: estimated } : {},
+    ...cacheReuseExpected !== void 0 ? { cacheReuseExpected } : {},
+    ...cacheMetricsLine ? { cacheMetricsLine } : {}
+  };
+}
+function estimateSystemTokensLite(systemMessages) {
+  let n = 0;
+  for (const m of systemMessages) n += 4 + Math.ceil((m.content ?? "").length / 4);
+  return n;
+}
+function estimateToolSchemaTokensLite(tools) {
+  let n = 0;
+  for (const t of tools) {
+    n += Math.ceil((t.name ?? "").length / 4);
+    n += Math.ceil((t.description ?? "").length / 4);
+    n += Math.ceil(JSON.stringify(t.parameters ?? {}).length / 4);
+  }
+  return n + tools.length * 4;
+}
+function estimateConversationTokensLite(messages) {
+  let n = 0;
+  for (const m of messages) {
+    n += 4 + Math.ceil((m.content ?? "").length / 4);
+    if (m.toolCalls) {
+      for (const tc of m.toolCalls) {
+        n += Math.ceil(tc.name.length / 4) + Math.ceil(tc.id.length / 4);
+        n += Math.ceil(JSON.stringify(tc.args ?? {}).length / 4);
+      }
+    }
+    if (m.reasoningContent) n += Math.ceil(m.reasoningContent.length / 4);
+    if (m.toolCallId) n += Math.ceil(m.toolCallId.length / 4);
+  }
+  return n;
+}
+var RESERVED_OUTPUT_TOKENS;
+var init_tokenBudget = __esm({
+  "src/cli/budget/tokenBudget.ts"() {
+    "use strict";
+    init_envNumber();
+    init_historyCompaction();
+    init_observationStore();
+    init_capabilities();
+    RESERVED_OUTPUT_TOKENS = 8192;
+  }
+});
+
+// src/cli/budget/retrievalPolicy.ts
+var retrievalPolicy_exports = {};
+__export(retrievalPolicy_exports, {
+  BASELINE_RETRIEVAL_POLICY: () => BASELINE_RETRIEVAL_POLICY,
+  RETRIEVAL_PRESSURE_THRESHOLDS: () => RETRIEVAL_PRESSURE_THRESHOLDS,
+  estimateHistoryOccupancy: () => estimateHistoryOccupancy,
+  filterSkillCatalogByBand: () => filterSkillCatalogByBand,
+  resolveRetrievalPolicy: () => resolveRetrievalPolicy,
+  retrievalBand: () => retrievalBand
+});
+function retrievalBand(occupancy) {
+  const occ = Number.isFinite(occupancy) ? Math.max(0, Math.min(1, occupancy)) : 0;
+  if (occ >= RETRIEVAL_PRESSURE_THRESHOLDS.high) return "high";
+  if (occ >= RETRIEVAL_PRESSURE_THRESHOLDS.medium) return "medium";
+  return "low";
+}
+function resolveRetrievalPolicy(occupancy) {
+  const band = retrievalBand(occupancy);
+  const packing = PRESSURE_PACKING[band];
+  const weights = PRESSURE_WEIGHTS[band];
+  return {
+    band,
+    ...packing,
+    ...weights ? { weights } : {}
+  };
+}
+function estimateHistoryOccupancy(history2, opts) {
+  const limit = resolveContextLimit(opts?.model, opts?.provider);
+  return Math.min(1, estimateHistoryTokens(history2) / limit);
+}
+function filterSkillCatalogByBand(skills, band) {
+  if (band !== "high") return [...skills];
+  return skills.filter((s) => s.estimatedCost !== "high");
+}
+var RETRIEVAL_PRESSURE_THRESHOLDS, BASELINE_RETRIEVAL_POLICY, PRESSURE_WEIGHTS, PRESSURE_PACKING;
+var init_retrievalPolicy = __esm({
+  "src/cli/budget/retrievalPolicy.ts"() {
+    "use strict";
+    init_tokenBudget();
+    RETRIEVAL_PRESSURE_THRESHOLDS = { medium: 0.5, high: 0.8 };
+    BASELINE_RETRIEVAL_POLICY = {
+      band: "low",
+      maxChars: 2e3,
+      maxMemories: 8
+    };
+    PRESSURE_WEIGHTS = {
+      low: void 0,
+      medium: {
+        semanticRelevance: 0.35,
+        lexicalRelevance: 0.2,
+        importance: 0.15,
+        confidence: 0.1,
+        recency: 0.05,
+        graphProximity: 0.1,
+        verificationBonus: 0.05
+      },
+      high: {
+        semanticRelevance: 0.4,
+        lexicalRelevance: 0.25,
+        importance: 0.15,
+        confidence: 0.1,
+        recency: 0,
+        graphProximity: 0.05,
+        verificationBonus: 0.05
+      }
+    };
+    PRESSURE_PACKING = {
+      low: { maxChars: 2e3, maxMemories: 8 },
+      medium: { maxChars: 1200, maxMemories: 6 },
+      high: { maxChars: 600, maxMemories: 4 }
+    };
+  }
+});
+
 // src/cli/tools/skillTool.ts
-function formatAvailableSkillsCatalog(cwd = process.cwd()) {
+function formatAvailableSkillsCatalog(cwd = process.cwd(), opts) {
   try {
     loadSkillMdSkills(cwd);
   } catch {
   }
-  const skills = listCodingSkills();
-  if (skills.length === 0) {
+  const band = opts?.pressureBand;
+  const all = listCodingSkills();
+  if (all.length === 0) {
     return "No skills registered.";
   }
-  const lines = skills.slice(0, 80).map((s) => {
+  const gated = band ? filterSkillCatalogByBand(all, band) : all;
+  const hidden = all.length - gated.length;
+  const lines = gated.slice(0, 80).map((s) => {
     const desc = (s.description || s.id).replace(/\s+/g, " ").trim().slice(0, 160);
     return `- ${s.id}: ${desc}`;
   });
+  if (hidden > 0) {
+    lines.push(
+      `(plus ${hidden} high-cost skill(s) hidden under context pressure \u2014 still loadable by name if truly needed)`
+    );
+  }
   return lines.join("\n");
 }
 function createSkillTool(opts) {
   const cwd = opts?.cwd ?? process.cwd();
-  const catalog = formatAvailableSkillsCatalog(cwd);
+  const catalog = formatAvailableSkillsCatalog(cwd, {
+    pressureBand: opts?.pressureBand
+  });
   return {
     name: "skill",
     description: "Load the full instructions for a named coding skill into this turn. Call when a skill matches the current task. Available skills:\n" + catalog,
@@ -40781,6 +41640,7 @@ var init_skillTool = __esm({
     init_skills2();
     init_toolTypes();
     init_skillsMd();
+    init_retrievalPolicy();
     SkillArgsSchema = external_exports.object({
       name: external_exports.string().min(1).describe("Skill id/name to load (from the available skills list).")
     });
@@ -43862,124 +44722,6 @@ var init_semantic2 = __esm({
   }
 });
 
-// src/cli/provider/capabilities.ts
-function frozenProfile(input) {
-  if (input.reasoning.levels) Object.freeze(input.reasoning.levels);
-  Object.freeze(input.reasoning);
-  Object.freeze(input.promptCache);
-  Object.freeze(input.toolCalling);
-  Object.freeze(input.buildRecovery);
-  Object.freeze(input.sampling);
-  Object.freeze(input.compaction);
-  return Object.freeze(input);
-}
-function resolveHarnessProfile(model, providerId) {
-  const provider = providerId?.trim().toLowerCase();
-  if (provider === "deepseek") return "deepseek-v4";
-  if (provider === "grok") return "grok";
-  if (provider === "minimax") return "minimax";
-  if (provider === "glm") return "glm";
-  if (model && DEEPSEEK_RE.test(model)) return "deepseek-v4";
-  if (model && GROK_RE.test(model)) return "grok";
-  if (model && MINIMAX_RE.test(model)) return "minimax";
-  if (model && GLM_RE.test(model)) return "glm";
-  return "default";
-}
-function capabilitiesFor(model, providerId) {
-  switch (resolveHarnessProfile(model, providerId)) {
-    case "deepseek-v4":
-      return DEEPSEEK_V4_CAPS;
-    case "grok":
-      return GROK_CAPS;
-    case "minimax":
-      return model && MINIMAX_M3_RE.test(model) ? MINIMAX_M3_CAPS : MINIMAX_M2_CAPS;
-    case "glm":
-      return GLM_CAPS;
-    default:
-      return DEFAULT_CAPS;
-  }
-}
-var SHARED_COMPACTION, DEFAULT_CAPS, DEEPSEEK_V4_CAPS, GROK_CAPS, MINIMAX_M3_CAPS, MINIMAX_M2_CAPS, GLM_CAPS, DEEPSEEK_RE, GROK_RE, MINIMAX_RE, MINIMAX_M3_RE, GLM_RE;
-var init_capabilities = __esm({
-  "src/cli/provider/capabilities.ts"() {
-    "use strict";
-    SHARED_COMPACTION = { warnAt: 0.7, compactAt: 0.85, hardAt: 0.95 };
-    DEFAULT_CAPS = frozenProfile({
-      contextWindow: 4e5,
-      reasoning: { supported: true, levels: ["low", "medium", "high"], replayReasoning: true },
-      promptCache: { supported: true, pricedCacheRead: false },
-      toolCalling: { parallel: true },
-      buildRecovery: { forceToolChoice: false, maxForcedTurns: 0 },
-      sampling: { temperature: 0.7 },
-      compaction: { ...SHARED_COMPACTION },
-      profile: "default"
-    });
-    DEEPSEEK_V4_CAPS = frozenProfile({
-      contextWindow: 1e6,
-      reasoning: { supported: true, levels: ["high", "max"], replayReasoning: true },
-      promptCache: { supported: true, pricedCacheRead: true },
-      toolCalling: { parallel: true },
-      buildRecovery: { forceToolChoice: false, maxForcedTurns: 0 },
-      sampling: { temperature: 0.7 },
-      compaction: { ...SHARED_COMPACTION },
-      profile: "deepseek-v4"
-    });
-    GROK_CAPS = frozenProfile({
-      contextWindow: 5e5,
-      reasoning: {
-        supported: true,
-        levels: ["low", "medium", "high", "xhigh"],
-        replayReasoning: false
-      },
-      promptCache: {
-        supported: true,
-        pricedCacheRead: true,
-        conversationAffinityHeader: "x-grok-conv-id"
-      },
-      toolCalling: { parallel: true },
-      buildRecovery: { forceToolChoice: true, maxForcedTurns: 1 },
-      sampling: { temperature: 0.7 },
-      compaction: { ...SHARED_COMPACTION },
-      profile: "grok"
-    });
-    MINIMAX_M3_CAPS = frozenProfile({
-      contextWindow: 1e6,
-      reasoning: { supported: true, replayReasoning: true },
-      promptCache: { supported: false, pricedCacheRead: false },
-      toolCalling: { parallel: true },
-      buildRecovery: { forceToolChoice: false, maxForcedTurns: 0 },
-      sampling: { temperature: 0.7 },
-      compaction: { ...SHARED_COMPACTION },
-      profile: "minimax"
-    });
-    MINIMAX_M2_CAPS = frozenProfile({
-      contextWindow: 204800,
-      reasoning: { supported: true, replayReasoning: true },
-      promptCache: { supported: false, pricedCacheRead: false },
-      toolCalling: { parallel: true },
-      buildRecovery: { forceToolChoice: false, maxForcedTurns: 0 },
-      sampling: { temperature: 0.7 },
-      compaction: { ...SHARED_COMPACTION },
-      profile: "minimax"
-    });
-    GLM_CAPS = frozenProfile({
-      contextWindow: 2e5,
-      reasoning: { supported: true, replayReasoning: true },
-      promptCache: { supported: true, pricedCacheRead: false },
-      toolCalling: { parallel: true },
-      buildRecovery: { forceToolChoice: false, maxForcedTurns: 0 },
-      sampling: { temperature: 0.7 },
-      compaction: { ...SHARED_COMPACTION },
-      profile: "glm"
-    });
-    DEEPSEEK_RE = /^deepseek-(?:v4(?:\.|-|$)|chat$|reasoner$)/i;
-    GROK_RE = /^grok(?:\.|-|$)/i;
-    MINIMAX_RE = /^minimax(?:\.|-|$)/i;
-    MINIMAX_M3_RE = /^minimax-m3(?:\.|-|$)/i;
-    GLM_RE = /^glm(?:\.|-|$)/i;
-  }
-});
-
 // src/cli/provider/openai-compatible.ts
 var openai_compatible_exports = {};
 __export(openai_compatible_exports, {
@@ -44163,6 +44905,12 @@ ${notes}
   }
   return { role: m.role, content: m.content };
 }
+function positiveEnvInt(name) {
+  const raw = process.env[name];
+  if (!raw) return void 0;
+  const n = Number.parseInt(raw, 10);
+  return Number.isFinite(n) && n > 0 ? n : void 0;
+}
 function openaiCompatibleProvider(config2) {
   return async function* (params) {
     const capabilities = capabilitiesFor(params.model, config2.providerId);
@@ -44195,6 +44943,8 @@ function openaiCompatibleProvider(config2) {
     };
     if (typeof generation?.maxTokens === "number" && generation.maxTokens > 0) {
       body.max_tokens = generation.maxTokens;
+    } else if (typeof capabilities.maxOutputTokens === "number" && capabilities.maxOutputTokens > 0) {
+      body.max_tokens = positiveEnvInt("ZELARI_MAX_OUTPUT_TOKENS") ?? capabilities.maxOutputTokens;
     }
     const thinkingSpec = config2.thinking ?? "auto";
     if (config2.providerId === "deepseek" && thinkingSpec === "auto") {
@@ -46386,8 +47136,8 @@ function defaultPermissionPolicy(overrides) {
   return {
     read: parseAction(process.env.ZELARI_PERMISSION_READ, "allow"),
     write: parseAction(process.env.ZELARI_PERMISSION_WRITE, "allow"),
-    execute: parseAction(process.env.ZELARI_PERMISSION_EXECUTE, "allow"),
-    network: parseAction(process.env.ZELARI_PERMISSION_NETWORK, "allow"),
+    execute: parseAction(process.env.ZELARI_PERMISSION_EXECUTE, "ask"),
+    network: parseAction(process.env.ZELARI_PERMISSION_NETWORK, "ask"),
     ui: "allow",
     auto: isAutoPermissions(),
     ...overrides
@@ -47353,7 +48103,7 @@ function matchResourceClaimLayered(layers, precedence, claim, root) {
 function resolveClaimsVerdict(layers, precedence, toolName, args, root) {
   const claims = resourceClaimsFor(toolName, args);
   const matchedRules = [];
-  if (!layers || claims.length === 0) return { matchedRules };
+  if (!layers || claims.length === 0) return { claims, matchedRules };
   const effects = [];
   for (const claim of claims) {
     const hit = matchResourceClaimLayered(layers, precedence, claim, root);
@@ -47362,7 +48112,26 @@ function resolveClaimsVerdict(layers, precedence, toolName, args, root) {
       matchedRules.push(hit);
     }
   }
-  return effects.length > 0 ? { effect: intersectEffects(...effects), matchedRules } : { matchedRules };
+  return effects.length > 0 ? { effect: intersectEffects(...effects), claims, matchedRules } : { claims, matchedRules };
+}
+function describeResourceClaim(claim) {
+  const cap3 = (s) => s.length > 160 ? `${s.slice(0, 157)}\u2026` : s;
+  switch (claim.kind) {
+    case "path":
+      return cap3(`path ${claim.operation}: ${claim.path}`);
+    case "process":
+      return cap3(`process: ${[claim.executable, ...claim.argv ?? []].join(" ")}`);
+    case "network":
+      return cap3(`network: ${claim.host}${claim.port !== void 0 ? `:${claim.port}` : ""}`);
+    case "mcp":
+      return cap3(`mcp: ${claim.server}/${claim.tool}`);
+    case "ssh":
+      return cap3(`ssh: ${claim.target}${claim.command ? ` (${claim.command})` : ""}`);
+    case "ui":
+      return cap3(`ui: ${claim.action}`);
+    case "agent":
+      return cap3(`agent: ${claim.role}`);
+  }
 }
 var MAX_RAW_SHELL_STRIP_DEPTH, RAW_SHELL_WRAPPERS, RAW_SHELL_INTERPRETERS;
 var init_resourceClaims = __esm({
@@ -47819,7 +48588,7 @@ function createBuiltinToolRegistry(options = {}) {
     registry4.register(withPerm(askUserTool));
   }
   const enableSkill = options.enableSkill !== false && options.readOnly !== true && !gauntletParent && profile !== "explore" && profile !== "verify";
-  const skillTool = enableSkill ? withPerm(createSkillTool({ cwd: root })) : null;
+  const skillTool = enableSkill ? withPerm(createSkillTool({ cwd: root, pressureBand: options.pressureBand })) : null;
   if (skillTool) {
     registry4.register(skillTool);
   }
@@ -48113,11 +48882,17 @@ function wrapWithPermissions(original, policy, onAsk, agentLayers, precedence =
           );
         }
         try {
+          const expanded = claims?.claims ?? resourceClaimsFor(original.name, input ?? {});
           const ok = await onAsk({
             toolName: original.name,
             reason: decision.reason,
             categories: decision.categories,
-            args: input
+            args: input,
+            policyNote: rulePrefix || void 0,
+            claims: expanded.slice(0, 6).map((c) => ({
+              kind: c.kind,
+              summary: describeResourceClaim(c)
+            }))
           });
           if (!ok) {
             return typedErr(
@@ -50136,15 +50911,39 @@ function spineMemoryEventNote(handle, event) {
   } catch {
   }
 }
+function bufferOf(holder) {
+  let buf = buffers.get(holder);
+  if (!buf) {
+    buf = [];
+    buffers.set(holder, buf);
+  }
+  return buf;
+}
 function memorySinkFor(holder) {
   return (event) => {
     const handle = holder.current;
-    if (handle) spineMemoryEventNote(handle, event);
+    if (handle) {
+      spineMemoryEventNote(handle, event);
+      return;
+    }
+    const buf = bufferOf(holder);
+    if (buf.length < PRE_BIND_BUFFER_CAP) buf.push(event);
+    else holder.droppedEvents = (holder.droppedEvents ?? 0) + 1;
   };
 }
+function flushMemorySpineNotes(holder) {
+  const handle = holder.current;
+  const buf = buffers.get(holder);
+  if (!handle || !buf?.length) return;
+  const pending = buf.splice(0, buf.length);
+  for (const event of pending) spineMemoryEventNote(handle, event);
+}
+var PRE_BIND_BUFFER_CAP, buffers;
 var init_spineTelemetry = __esm({
   "src/cli/memory/spineTelemetry.ts"() {
     "use strict";
+    PRE_BIND_BUFFER_CAP = 32;
+    buffers = /* @__PURE__ */ new WeakMap();
   }
 });
 
@@ -50356,567 +51155,123 @@ var init_fileStateStore = __esm({
           mode: input.mode,
           label: input.label,
           verification: input.verification,
-          changedPaths: input.changedPaths ?? [],
-          discoveryCount: input.discoveries?.length ?? 0
-        };
-      }
-      async head() {
-        return null;
-      }
-      async get() {
-        return null;
-      }
-      async list() {
-        return [];
-      }
-      async setHead(id3) {
-        throw new Error(`NoopDurableStateStore.setHead: state disabled (id=${id3})`);
-      }
-      async loadDiscoveries() {
-        return [];
-      }
-      async materializeContext() {
-        return "";
-      }
-      async close() {
-      }
-    };
-  }
-});
-
-// src/cli/hooks/messageHelpers.ts
-function appendSystem(setMessages, content, ts = Date.now()) {
-  setMessages((prev2) => [
-    ...prev2,
-    { id: crypto.randomUUID(), role: "system", content, ts }
-  ]);
-}
-function appendUser(setMessages, content, ts = Date.now()) {
-  setMessages((prev2) => [
-    ...prev2,
-    { id: crypto.randomUUID(), role: "user", content, ts }
-  ]);
-}
-function appendOrExtendStreamingAssistant(setMessages, fullContent, ts, memberContext) {
-  setMessages((prev2) => {
-    const last = prev2[prev2.length - 1];
-    if (last && last.role === "assistant" && last.id.startsWith("streaming-")) {
-      return [...prev2.slice(0, -1), { ...last, content: fullContent }];
-    }
-    return [
-      ...prev2,
-      {
-        id: `streaming-${crypto.randomUUID()}`,
-        role: "assistant",
-        content: fullContent,
-        ts,
-        ...memberContext?.memberId ? { memberId: memberContext.memberId } : {},
-        ...memberContext?.memberName ? { memberName: memberContext.memberName } : {}
-      }
-    ];
-  });
-}
-function appendToolStart(setMessages, toolName, toolCallId, args, ts) {
-  const argsPreview = formatToolSummary(toolName, args);
-  setMessages((prev2) => [
-    ...prev2,
-    {
-      id: crypto.randomUUID(),
-      role: "tool",
-      content: argsPreview,
-      ts,
-      toolName,
-      toolCallId,
-      toolOk: void 0,
-      toolDurationMs: void 0
-    }
-  ]);
-}
-function updateToolMessageEnd(setMessages, toolCallId, isError, durationMs, result) {
-  setMessages((prev2) => {
-    for (let i = prev2.length - 1; i >= 0; i--) {
-      const m = prev2[i];
-      if (m && m.role === "tool" && m.toolCallId === toolCallId && m.toolDurationMs === void 0) {
-        const updated = [...prev2];
-        updated[i] = {
-          ...m,
-          toolOk: !isError,
-          toolDurationMs: durationMs,
-          ...result !== void 0 ? {
-            toolResult: toolResultForStorage(
-              m.toolName ?? "",
-              result,
-              isError
-            )
-          } : {}
-        };
-        return updated;
-      }
-    }
-    return prev2;
-  });
-}
-function finalizeStreamingAssistant(setMessages) {
-  setMessages((prev2) => {
-    const last = prev2[prev2.length - 1];
-    if (last && last.role === "assistant" && last.id.startsWith("streaming-")) {
-      return [
-        ...prev2.slice(0, -1),
-        { ...last, id: last.id.slice("streaming-".length) }
-      ];
-    }
-    return prev2;
-  });
-}
-var init_messageHelpers = __esm({
-  "src/cli/hooks/messageHelpers.ts"() {
-    "use strict";
-    init_toolFormat();
-    init_toolFormat();
-  }
-});
-
-// src/cli/budget/historySummary.ts
-function extractiveHistorySummary(dropped, opts) {
-  const maxChars = opts?.maxChars ?? MAX_SUMMARY_CHARS;
-  if (dropped.length === 0) return "No prior turns.";
-  const userGoals = [];
-  const assistantNotes = [];
-  const userConstraints = [];
-  const unresolved = [];
-  const verification = [];
-  const decisions = [];
-  const tools = /* @__PURE__ */ new Map();
-  const files = /* @__PURE__ */ new Set();
-  let toolResults = 0;
-  for (const m of dropped) {
-    if (m.role === "user" && m.content.trim()) {
-      const goal = oneLine(m.content, 220);
-      userGoals.push(goal);
-      if (/\b(must|never|required|only|do not|constraint|vincolo|deve|senza)\b/i.test(goal)) userConstraints.push(goal);
-    } else if (m.role === "assistant") {
-      if (m.content.trim()) {
-        const note = oneLine(m.content, 220);
-        assistantNotes.push(note);
-        if (/\b(fail|failed|error|unresolved|remaining|todo|blocked|gap|errore|fallit|irrisolt|manca)\b/i.test(note)) unresolved.push(note);
-        if (/\b(test|typecheck|build|verify|verification|passed|failed|green|red)\b/i.test(note)) verification.push(note);
-        if (/\b(decid|decision|chosen|choose|scelt|adopt|implement)\b/i.test(note)) decisions.push(note);
-      }
-      if (m.toolCalls) {
-        for (const tc of m.toolCalls) {
-          tools.set(tc.name, (tools.get(tc.name) ?? 0) + 1);
-          collectPaths3(tc.args, files);
-        }
-      }
-    } else if (m.role === "tool") {
-      toolResults += 1;
-      collectPathsFromText(m.content, files);
-      if (/\b(fail|failed|error|exception|blocked|errore|fallit)\b/i.test(m.content)) unresolved.push(oneLine(m.content, 220));
-      if (/\b(test|typecheck|build|verify|passed|failed|success)\b/i.test(m.content)) verification.push(oneLine(m.content, 220));
-    }
-  }
-  const parts = [
-    "[history-summary] Earlier turns were compacted to stay within the context budget.",
-    `Dropped ${dropped.length} message(s) (${userGoals.length} user, ${assistantNotes.length} assistant notes, ${toolResults} tool results).`
-  ];
-  if (userGoals.length) {
-    parts.push("## User goals / requests");
-    for (const g of userGoals.slice(-6)) parts.push(`- ${g}`);
-  }
-  if (userConstraints.length) {
-    parts.push("## User constraints (preserve exactly)");
-    for (const item of [...new Set(userConstraints)].slice(-8)) parts.push(`- ${item}`);
-  }
-  if (unresolved.length) {
-    parts.push("## Unresolved failures / pending repair");
-    for (const item of [...new Set(unresolved)].slice(-8)) parts.push(`- ${item}`);
-  }
-  if (verification.length) {
-    parts.push("## Latest verification state");
-    for (const item of [...new Set(verification)].slice(-6)) parts.push(`- ${item}`);
-  }
-  if (decisions.length) {
-    parts.push("## Recent active decisions");
-    for (const item of [...new Set(decisions)].slice(-6)) parts.push(`- ${item}`);
-  }
-  if (assistantNotes.length) {
-    parts.push("## Assistant conclusions (truncated)");
-    for (const a of assistantNotes.slice(-5)) parts.push(`- ${a}`);
-  }
-  if (tools.size) {
-    const ranked = [...tools.entries()].sort((a, b) => b[1] - a[1]).slice(0, 12).map(([n, c]) => `${n}\xD7${c}`);
-    parts.push(`## Tools used: ${ranked.join(", ")}`);
-  }
-  if (files.size) {
-    const list = [...files].slice(0, 24);
-    parts.push(`## Paths mentioned: ${list.join(", ")}`);
-  }
-  parts.push(
-    "Continue from the recent messages below; do not re-ask goals already answered above unless the user changes them."
-  );
-  let out = parts.join("\n");
-  if (out.length > maxChars) {
-    out = `${out.slice(0, maxChars - 1)}\u2026`;
-  }
-  return out;
-}
-function oneLine(s, max) {
-  const t = s.replace(/\s+/g, " ").trim();
-  if (t.length <= max) return t;
-  return `${t.slice(0, max - 1)}\u2026`;
-}
-function collectPaths3(args, out) {
-  if (!args || typeof args !== "object") return;
-  const obj = args;
-  for (const key of ["path", "file", "filepath", "filePath", "target", "cwd"]) {
-    const v = obj[key];
-    if (typeof v === "string" && v.length > 1 && v.length < 260) {
-      out.add(v.replace(/\\/g, "/"));
-    }
-  }
-  if (typeof obj.file_path === "string") out.add(String(obj.file_path));
-}
-function collectPathsFromText(text, out) {
-  const re = /(?:^|[\s"'`])((?:[\w.-]+\/)+[\w.-]+\.\w{1,8})/g;
-  let m;
-  let n = 0;
-  while ((m = re.exec(text)) !== null && n < 8) {
-    out.add(m[1]);
-    n += 1;
-  }
-}
-var MAX_SUMMARY_CHARS;
-var init_historySummary = __esm({
-  "src/cli/budget/historySummary.ts"() {
-    "use strict";
-    MAX_SUMMARY_CHARS = 3500;
-  }
-});
-
-// src/cli/budget/llmCompact.ts
-function isLlmCompactEnabled() {
-  const v = process.env.ZELARI_LLM_COMPACT?.trim().toLowerCase();
-  if (v === "0" || v === "false" || v === "off" || v === "no") return false;
-  return true;
-}
-function compactModelOverride() {
-  const v = process.env.ZELARI_COMPACT_MODEL?.trim();
-  return v ? v : void 0;
-}
-async function llmSummarizeHistoryReplay(input) {
-  const override = input.overrideModel ?? compactModelOverride();
-  const model = override ?? input.model;
-  const cacheReuseExpected = !override;
-  if (!isLlmCompactEnabled()) return { summary: null, model, cacheReuseExpected };
-  if (input.droppedMessages.length === 0) {
-    return { summary: null, model, cacheReuseExpected };
-  }
-  const messages = [
-    ...input.systemMessages,
-    ...input.droppedMessages,
-    {
-      role: "user",
-      content: COMPACTION_INSTRUCTION
-    }
-  ];
-  const controller = new AbortController();
-  const timeout = setTimeout(() => controller.abort(), REPLAY_TIMEOUT_MS);
-  const onOuterAbort = () => controller.abort();
-  input.signal?.addEventListener("abort", onOuterAbort, { once: true });
-  try {
-    let text = "";
-    let emittedToolCall = false;
-    for await (const delta of input.providerStream({
-      provider: input.provider,
-      model,
-      messages,
-      // Tools stay advertised: dropping them would change the prefix token
-      // sequence and destroy cache reuse (explicit DSH decision). They are
-      // sorted canonically (same discipline as the live routed request and
-      // the snapshot fingerprints) so the replay prefix is byte-identical.
-      tools: [...input.tools].sort((a, b) => a.name < b.name ? -1 : a.name > b.name ? 1 : 0),
-      signal: controller.signal,
-      generation: {
-        purpose: "compaction",
-        temperature: 0.1,
-        maxTokens: 900
-      }
-    })) {
-      if (delta.kind === "text") text += delta.delta;
-      if (delta.kind === "tool_call") emittedToolCall = true;
-    }
-    if (emittedToolCall) return { summary: null, model, cacheReuseExpected };
-    if (!text.trim()) return { summary: null, model, cacheReuseExpected };
-    return { summary: text.trim(), model, cacheReuseExpected };
-  } catch {
-    return { summary: null, model, cacheReuseExpected };
-  } finally {
-    clearTimeout(timeout);
-    input.signal?.removeEventListener("abort", onOuterAbort);
-  }
-}
-var COMPACTION_INSTRUCTION, REPLAY_TIMEOUT_MS;
-var init_llmCompact = __esm({
-  "src/cli/budget/llmCompact.ts"() {
-    "use strict";
-    COMPACTION_INSTRUCTION = `
-You are now acting as a compaction engine for this coding-agent session.
-
-Condense the conversation ABOVE into a compact checkpoint sufficient to continue the task.
-
-Preserve:
-- user's goal and evolving intent
-- decisions already made
-- exact file paths and identifiers
-- code changes already completed
-- commands/errors that still matter
-- constraints
-- unfinished work
-- the single most likely next action
-
-Do not call tools.
-Do not mention this summarization request.
-Output only the checkpoint.
-Be concise.
-`.trim();
-    REPLAY_TIMEOUT_MS = 6e4;
-  }
-});
-
-// src/cli/hooks/historyCompaction.ts
-function compactedRangeFromDropped(dropped) {
-  if (dropped.length === 0) return void 0;
-  const seqs = [];
-  const sources = [];
-  for (const m of dropped) {
-    const hasCompactRange = typeof m.compactedFromSeq === "number" && Number.isInteger(m.compactedFromSeq) && m.compactedFromSeq > 0 && typeof m.compactedToSeq === "number" && Number.isInteger(m.compactedToSeq) && m.compactedToSeq >= m.compactedFromSeq;
-    if (hasCompactRange) {
-      seqs.push(m.compactedFromSeq, m.compactedToSeq);
-      sources.push(...m.sourceEventSeqs ?? []);
-      if (typeof m.seq === "number" && Number.isInteger(m.seq) && m.seq > 0) {
-        seqs.push(m.seq);
-        sources.push(m.seq);
-      }
-      continue;
-    }
-    if (typeof m.seq !== "number" || !Number.isInteger(m.seq) || m.seq < 1) return void 0;
-    seqs.push(m.seq);
-    sources.push(m.seq);
-  }
-  return {
-    fromSeq: Math.min(...seqs),
-    toSeq: Math.max(...seqs),
-    sourceEventSeqs: [...new Set(sources)]
-  };
-}
-function withDroppedRange(result, dropped, strategy) {
-  const range = compactedRangeFromDropped(dropped);
-  if (!range) return { ...result, strategy };
-  return { ...result, ...range, strategy };
-}
-function resolveMaxMessages(opts) {
-  const envTurns = envNumber(process.env.ZELARI_HISTORY_TURNS, { default: 6, min: 0 });
-  let turns = opts?.maxMessages ? Math.ceil(opts.maxMessages / 4) : envTurns;
-  if (opts?.durableStatePresent && !opts?.maxMessages && process.env.ZELARI_HISTORY_TURNS === void 0) {
-    turns = Math.min(turns, 3);
-  }
-  if (turns <= 0) return 0;
-  if (opts?.force && opts?.maxMessages) return Math.max(1, opts.maxMessages);
-  return turns * 4;
-}
-function findValidCutIndex(messages, naiveCut) {
-  let cut = naiveCut;
-  while (cut < messages.length) {
-    const kept = messages.slice(cut);
-    const declared = /* @__PURE__ */ new Set();
-    for (const m of kept) {
-      if (m.role === "assistant" && m.toolCalls) {
-        for (const tc of m.toolCalls) declared.add(tc.id);
+          changedPaths: input.changedPaths ?? [],
+          discoveryCount: input.discoveries?.length ?? 0
+        };
       }
-    }
-    let moved = false;
-    for (let k = 0; k < kept.length; k++) {
-      const m = kept[k];
-      if (m.role === "tool" && m.toolCallId && !declared.has(m.toolCallId)) {
-        for (let j = cut - 1; j >= 0; j--) {
-          const prev2 = messages[j];
-          if (prev2.role === "assistant" && prev2.toolCalls && prev2.toolCalls.some((tc) => tc.id === m.toolCallId)) {
-            cut = j;
-            moved = true;
-            break;
-          }
-        }
-        break;
+      async head() {
+        return null;
       }
-    }
-    if (!moved) break;
+      async get() {
+        return null;
+      }
+      async list() {
+        return [];
+      }
+      async setHead(id3) {
+        throw new Error(`NoopDurableStateStore.setHead: state disabled (id=${id3})`);
+      }
+      async loadDiscoveries() {
+        return [];
+      }
+      async materializeContext() {
+        return "";
+      }
+      async close() {
+      }
+    };
   }
-  return cut;
+});
+
+// src/cli/hooks/messageHelpers.ts
+function appendSystem(setMessages, content, ts = Date.now()) {
+  setMessages((prev2) => [
+    ...prev2,
+    { id: crypto.randomUUID(), role: "system", content, ts }
+  ]);
 }
-function resolvePruneLimits(opts) {
-  const maxChars = opts?.maxChars ?? envNumber(process.env.ZELARI_TOOL_RESULT_MAX_CHARS, { default: 8e3, min: 256 });
-  const rawTail = opts?.tailChars ?? envNumber(process.env.ZELARI_TOOL_RESULT_TAIL_CHARS, { default: 1e3, min: 0 });
-  const tailChars = Math.min(rawTail, maxChars);
-  return { maxChars, tailChars };
+function appendUser(setMessages, content, ts = Date.now()) {
+  setMessages((prev2) => [
+    ...prev2,
+    { id: crypto.randomUUID(), role: "user", content, ts }
+  ]);
 }
-function pruneToolResultsDetailed(messages, opts) {
-  const { maxChars, tailChars } = resolvePruneLimits(opts);
-  const headChars = maxChars - tailChars;
-  const stats = { pruned: 0, charsOmitted: 0 };
-  let changed = false;
-  const out = messages.map((m) => {
-    if (m.role !== "tool") return m;
-    const body = m.content ?? "";
-    if (body.length <= maxChars) return m;
-    const head = headChars > 0 ? body.slice(0, headChars) : "";
-    const tail2 = tailChars > 0 ? body.slice(-tailChars) : "";
-    const omitted = body.length - head.length - tail2.length;
-    changed = true;
-    stats.pruned += 1;
-    stats.charsOmitted += omitted;
-    return {
-      ...m,
-      content: [head, "\u2026[pruned " + omitted + " chars]\u2026", tail2].join(String.fromCharCode(10))
-    };
+function appendOrExtendStreamingAssistant(setMessages, fullContent, ts, memberContext) {
+  setMessages((prev2) => {
+    const last = prev2[prev2.length - 1];
+    if (last && last.role === "assistant" && last.id.startsWith("streaming-")) {
+      return [...prev2.slice(0, -1), { ...last, content: fullContent }];
+    }
+    return [
+      ...prev2,
+      {
+        id: `streaming-${crypto.randomUUID()}`,
+        role: "assistant",
+        content: fullContent,
+        ts,
+        ...memberContext?.memberId ? { memberId: memberContext.memberId } : {},
+        ...memberContext?.memberName ? { memberName: memberContext.memberName } : {}
+      }
+    ];
   });
-  return {
-    messages: changed ? out : messages,
-    stats
-  };
-}
-function compactHistory(messages, opts) {
-  return compactHistoryDetailed(messages, opts).messages;
-}
-function buildCheckpointMessage(summaryText, range) {
-  return {
-    role: "user",
-    content: CHECKPOINT_WRAPPER_PREFIX + "\n\n<compacted-summary>\n" + summaryText + "\n</compacted-summary>",
-    ...range ? {
-      compactedFromSeq: range.fromSeq,
-      compactedToSeq: range.toSeq,
-      sourceEventSeqs: [...range.sourceEventSeqs]
-    } : {}
-  };
 }
-function compactHistoryDetailed(messages, opts) {
-  const maxMessages = resolveMaxMessages(opts);
-  if (maxMessages === 0) {
-    return withDroppedRange(
-      { messages: [], compacted: true, messagesRemoved: messages.length, summary: "" },
-      messages,
-      "extractive"
-    );
-  }
-  if (messages.length <= maxMessages * 2 && !opts?.force) {
-    return {
-      messages,
-      compacted: false,
-      messagesRemoved: 0,
-      summary: ""
-    };
-  }
-  const naiveCut = Math.max(0, messages.length - maxMessages);
-  const cut = findValidCutIndex(messages, naiveCut);
-  if (cut === 0) {
-    return {
-      messages,
-      compacted: false,
-      messagesRemoved: 0,
-      summary: ""
-    };
-  }
-  const droppedMsgs = messages.slice(0, cut);
-  const droppedRange = compactedRangeFromDropped(droppedMsgs);
-  const pruned = pruneToolResultsDetailed(messages.slice(cut));
-  const kept = pruned.messages;
-  const summaryText = extractiveHistorySummary(droppedMsgs);
-  const summary = buildCheckpointMessage(
-    summaryText || `${COMPACT_MARKER} ${cut} earlier message(s) dropped.`,
-    droppedRange
-  );
-  return withDroppedRange(
+function appendToolStart(setMessages, toolName, toolCallId, args, ts) {
+  const argsPreview = formatToolSummary(toolName, args);
+  setMessages((prev2) => [
+    ...prev2,
     {
-      messages: [summary, ...kept],
-      compacted: true,
-      messagesRemoved: cut,
-      summary: summary.content,
-      prunedToolResults: pruned.stats.pruned
-    },
-    droppedMsgs,
-    "extractive"
-  );
+      id: crypto.randomUUID(),
+      role: "tool",
+      content: argsPreview,
+      ts,
+      toolName,
+      toolCallId,
+      toolOk: void 0,
+      toolDurationMs: void 0
+    }
+  ]);
 }
-async function compactHistoryAsync(messages, opts) {
-  const base2 = compactHistoryDetailed(messages, opts);
-  if (!base2.compacted || base2.messagesRemoved === 0) return base2;
-  const cut = base2.messagesRemoved;
-  const droppedMsgs = messages.slice(0, cut);
-  const extractive = extractiveHistorySummary(droppedMsgs);
-  let summaryText = extractive;
-  let cacheReuseExpected;
-  let replayExactPrefix;
-  const canReplay = !!(opts?.providerStream && opts?.requestSnapshot);
-  if (canReplay) {
-    try {
-      const replay = await llmSummarizeHistoryReplay({
-        providerStream: opts.providerStream,
-        provider: opts.requestSnapshot.provider,
-        model: opts.requestSnapshot.model,
-        systemMessages: opts.requestSnapshot.systemMessages,
-        tools: opts.requestSnapshot.tools,
-        droppedMessages: droppedMsgs,
-        signal: opts?.signal
-      });
-      cacheReuseExpected = replay.cacheReuseExpected;
-      if (replay.summary && replay.summary.trim().length > 40) {
-        const sourceTokens = roughTokens(droppedMsgs);
-        const summaryTok = Math.ceil(replay.summary.length / 4);
-        if (summaryTok < sourceTokens) {
-          summaryText = replay.summary.trim();
-        }
+function updateToolMessageEnd(setMessages, toolCallId, isError, durationMs, result) {
+  setMessages((prev2) => {
+    for (let i = prev2.length - 1; i >= 0; i--) {
+      const m = prev2[i];
+      if (m && m.role === "tool" && m.toolCallId === toolCallId && m.toolDurationMs === void 0) {
+        const updated = [...prev2];
+        updated[i] = {
+          ...m,
+          toolOk: !isError,
+          toolDurationMs: durationMs,
+          ...result !== void 0 ? {
+            toolResult: toolResultForStorage(
+              m.toolName ?? "",
+              result,
+              isError
+            )
+          } : {}
+        };
+        return updated;
       }
-    } catch {
     }
-  }
-  const pruned = pruneToolResultsDetailed(messages.slice(cut));
-  const kept = pruned.messages;
-  const summary = buildCheckpointMessage(summaryText, compactedRangeFromDropped(droppedMsgs));
-  const usedLlm = summaryText !== extractive && summaryText.trim().length > 40;
-  return withDroppedRange(
-    {
-      messages: [summary, ...kept],
-      compacted: true,
-      messagesRemoved: cut,
-      summary: summaryText,
-      prunedToolResults: pruned.stats.pruned,
-      cacheReuseExpected,
-      replayExactPrefix
-    },
-    droppedMsgs,
-    usedLlm ? "llm" : "extractive"
-  );
+    return prev2;
+  });
 }
-function roughTokens(msgs) {
-  let n = 0;
-  for (const m of msgs) {
-    n += Math.ceil((m.content ?? "").length / 4);
-    if (m.toolCalls) {
-      for (const tc of m.toolCalls) {
-        n += Math.ceil(JSON.stringify(tc.args ?? {}).length / 4);
-      }
+function finalizeStreamingAssistant(setMessages) {
+  setMessages((prev2) => {
+    const last = prev2[prev2.length - 1];
+    if (last && last.role === "assistant" && last.id.startsWith("streaming-")) {
+      return [
+        ...prev2.slice(0, -1),
+        { ...last, id: last.id.slice("streaming-".length) }
+      ];
     }
-  }
-  return Math.max(1, n);
+    return prev2;
+  });
 }
-var COMPACT_MARKER, CHECKPOINT_WRAPPER_PREFIX;
-var init_historyCompaction = __esm({
-  "src/cli/hooks/historyCompaction.ts"() {
+var init_messageHelpers = __esm({
+  "src/cli/hooks/messageHelpers.ts"() {
     "use strict";
-    init_historySummary();
-    init_llmCompact();
-    init_envNumber();
-    COMPACT_MARKER = "[history] Earlier turns were compacted to stay within the context budget.";
-    CHECKPOINT_WRAPPER_PREFIX = "This is an automatically generated checkpoint of earlier conversation. Treat it as established context and continue directly.";
+    init_toolFormat();
+    init_toolFormat();
   }
 });
 
@@ -51229,215 +51584,6 @@ var init_phase = __esm({
   }
 });
 
-// src/cli/budget/tokenBudget.ts
-function mergeCompactRange(into, r) {
-  if (r.fromSeq !== void 0 && r.toSeq !== void 0) {
-    into.fromSeq = into.fromSeq === void 0 ? r.fromSeq : Math.min(into.fromSeq, r.fromSeq);
-    into.toSeq = into.toSeq === void 0 ? r.toSeq : Math.max(into.toSeq, r.toSeq);
-    if (r.sourceEventSeqs) into.sourceSeqs.push(...r.sourceEventSeqs);
-  }
-  if (r.strategy === "llm") into.strategy = "llm";
-  else if (r.strategy && !into.strategy) into.strategy = r.strategy;
-}
-function estimateTokens2(text) {
-  if (!text) return 0;
-  return Math.max(1, Math.ceil(text.length / 4));
-}
-function estimateHistoryTokens(messages) {
-  let n = 0;
-  for (const m of messages) {
-    n += estimateTokens2(m.content);
-    if (m.toolCalls) {
-      for (const tc of m.toolCalls) {
-        n += estimateTokens2(tc.name) + estimateTokens2(JSON.stringify(tc.args ?? {}));
-      }
-    }
-  }
-  return n;
-}
-function defaultContextLimitForModel(model, provider) {
-  return capabilitiesFor(model, provider).contextWindow;
-}
-function resolveContextLimit(model, provider) {
-  return envNumber(process.env.ZELARI_CONTEXT_LIMIT, {
-    default: defaultContextLimitForModel(model, provider),
-    min: 4e3,
-    max: 2e6
-  });
-}
-function phaseKnobs(phase2) {
-  return {
-    historyTurns: phase2 === "plan" ? envNumber(process.env.ZELARI_HISTORY_TURNS, { default: 8, min: 0 }) : envNumber(process.env.ZELARI_HISTORY_TURNS, { default: 6, min: 0 }),
-    maxToolLoopIterations: phase2 === "plan" ? envNumber(process.env.ZELARI_MAX_TOOL_LOOP_ITERATIONS, { default: 60, min: 1 }) : envNumber(process.env.ZELARI_MAX_TOOL_LOOP_ITERATIONS, {
-      default: 120,
-      min: 1
-    })
-  };
-}
-async function applyBudgetPolicyAsync(history2, phase2, opts) {
-  const contextLimit = resolveContextLimit(opts?.model, opts?.provider);
-  const compact3 = capabilitiesFor(opts?.model, opts?.provider).compaction;
-  const sessionExtra = opts?.sessionTokens ?? 0;
-  const warnings = [];
-  let { historyTurns, maxToolLoopIterations } = phaseKnobs(phase2);
-  const envelope = opts?.requestSnapshot ?? null;
-  const surface = envelope?.snapshot ?? opts?.requestSurface ?? null;
-  const replayBase = surface ? {
-    provider: surface.provider,
-    model: surface.model,
-    systemMessages: surface.systemMessages,
-    tools: surface.tools
-  } : opts?.providerStream ? { provider: "local", model: opts?.model ?? "unknown", systemMessages: [], tools: [] } : null;
-  const headerTokens = surface ? estimateSystemTokensLite(surface.systemMessages) + estimateToolSchemaTokensLite(surface.tools) : 0;
-  const convTokensOf = (h) => estimateConversationTokensLite(h);
-  let hist = history2;
-  let estimated = surface ? headerTokens + convTokensOf(hist) + RESERVED_OUTPUT_TOKENS : estimateHistoryTokens(hist) + sessionExtra;
-  let occupancy = Math.min(1, estimated / contextLimit);
-  let compactSummary = "";
-  let messagesRemoved = 0;
-  let cacheReuseExpected;
-  let prunedTotal = 0;
-  const compactRange = { sourceSeqs: [] };
-  if (occupancy >= compact3.warnAt && occupancy < compact3.compactAt) {
-    warnings.push(
-      `[budget] context ~${Math.round(occupancy * 100)}% full (${estimated}/${contextLimit} tok full-request est.) \u2014 consider /compact or shorter replies.`
-    );
-  }
-  if (occupancy >= 0.8) {
-    const pruned = pruneToolResultsDetailed(hist);
-    if (pruned.stats.pruned > 0) {
-      hist = pruned.messages;
-      prunedTotal += pruned.stats.pruned;
-      estimated = surface ? headerTokens + convTokensOf(hist) + RESERVED_OUTPUT_TOKENS : estimateHistoryTokens(hist) + sessionExtra;
-      occupancy = Math.min(1, estimated / contextLimit);
-      warnings.push(
-        `[budget] pruned ${pruned.stats.pruned} oversized tool result(s) \u2192 ${Math.round(occupancy * 100)}% (${estimated} tok).`
-      );
-    }
-  }
-  const fold = (r, label, forcedTurns) => {
-    hist = applySessionSurface(r.messages);
-    if (r.compacted) {
-      messagesRemoved += r.messagesRemoved;
-      if (r.summary) compactSummary = r.summary;
-      mergeCompactRange(compactRange, r);
-      if (r.cacheReuseExpected !== void 0) {
-        cacheReuseExpected = r.cacheReuseExpected;
-      }
-    }
-    estimated = surface ? headerTokens + convTokensOf(hist) + RESERVED_OUTPUT_TOKENS : estimateHistoryTokens(hist) + sessionExtra;
-    occupancy = Math.min(1, estimated / contextLimit);
-    warnings.push(
-      `[budget] ${label} \u2014 kept ~${forcedTurns} turns (${estimated} tok est.` + (r.messagesRemoved ? `, removed ${r.messagesRemoved} msgs` : "") + (r.cacheReuseExpected === false ? ", cache reuse NOT expected (model override)" : "") + ")."
-    );
-  };
-  if (occupancy >= compact3.compactAt) {
-    const forcedTurns = Math.max(1, Math.floor(historyTurns / 2));
-    historyTurns = forcedTurns;
-    maxToolLoopIterations = Math.min(
-      maxToolLoopIterations,
-      phase2 === "plan" ? 24 : 40
-    );
-    let r = await compactHistoryAsync(hist, {
-      maxMessages: Math.max(2, forcedTurns * 4),
-      force: true,
-      signal: opts?.signal,
-      ...replayBase ? { requestSnapshot: replayBase, providerStream: opts?.providerStream } : {}
-    });
-    if (!r.compacted) {
-      r = await compactHistoryAsync(hist, {
-        maxMessages: 2,
-        force: true,
-        signal: opts?.signal,
-        ...replayBase ? { requestSnapshot: replayBase, providerStream: opts?.providerStream } : {}
-      });
-    }
-    const label = r.cacheReuseExpected === false ? "llm-compact (model override)" : "auto-compact";
-    fold(r, label, forcedTurns);
-  }
-  if (occupancy >= compact3.hardAt) {
-    const hard = await compactHistoryAsync(hist, {
-      maxMessages: 2,
-      force: true,
-      signal: opts?.signal
-    });
-    fold(hard, hard.summary.includes("\xB7 llm") ? "llm-compact" : "HARD trim", 2);
-    historyTurns = 2;
-    maxToolLoopIterations = Math.min(maxToolLoopIterations, 16);
-    warnings.push(
-      "[budget] HARD context pressure (\u226595%) \u2014 prefer /clear or a new session if quality drops."
-    );
-  }
-  const cacheMetricsLine = envelope ? [
-    "compaction meter:",
-    `provider/model: ${envelope.snapshot.provider}/${envelope.snapshot.model}`,
-    `headerFingerprint: ${envelope.snapshot.headerFingerprint.slice(0, 12)}`,
-    `occupancy: ${Math.round(occupancy * 100)}% (${estimated}/${contextLimit})`,
-    ...envelope.usage?.cachedPromptTokens !== void 0 ? [`cachedPromptTokens: ${envelope.usage.cachedPromptTokens}`] : [],
-    ...cacheReuseExpected !== void 0 ? [`cacheReuseExpected: ${cacheReuseExpected}`] : []
-  ].join(" | ") : void 0;
-  return {
-    history: hist,
-    warnings,
-    maxToolLoopIterations,
-    historyTurns,
-    estimatedHistoryTokens: surface ? convTokensOf(hist) : estimated,
-    contextLimit,
-    occupancy,
-    compactSummary: compactSummary || void 0,
-    messagesRemoved: messagesRemoved || void 0,
-    ...compactRange.fromSeq !== void 0 && compactRange.toSeq !== void 0 ? {
-      compactedFromSeq: compactRange.fromSeq,
-      compactedToSeq: compactRange.toSeq,
-      compactSourceSeqs: compactRange.sourceSeqs,
-      compactStrategy: compactRange.strategy
-    } : {},
-    ...surface ? { contextPressureTokens: estimated } : {},
-    ...cacheReuseExpected !== void 0 ? { cacheReuseExpected } : {},
-    ...cacheMetricsLine ? { cacheMetricsLine } : {}
-  };
-}
-function estimateSystemTokensLite(systemMessages) {
-  let n = 0;
-  for (const m of systemMessages) n += 4 + Math.ceil((m.content ?? "").length / 4);
-  return n;
-}
-function estimateToolSchemaTokensLite(tools) {
-  let n = 0;
-  for (const t of tools) {
-    n += Math.ceil((t.name ?? "").length / 4);
-    n += Math.ceil((t.description ?? "").length / 4);
-    n += Math.ceil(JSON.stringify(t.parameters ?? {}).length / 4);
-  }
-  return n + tools.length * 4;
-}
-function estimateConversationTokensLite(messages) {
-  let n = 0;
-  for (const m of messages) {
-    n += 4 + Math.ceil((m.content ?? "").length / 4);
-    if (m.toolCalls) {
-      for (const tc of m.toolCalls) {
-        n += Math.ceil(tc.name.length / 4) + Math.ceil(tc.id.length / 4);
-        n += Math.ceil(JSON.stringify(tc.args ?? {}).length / 4);
-      }
-    }
-    if (m.reasoningContent) n += Math.ceil(m.reasoningContent.length / 4);
-    if (m.toolCallId) n += Math.ceil(m.toolCallId.length / 4);
-  }
-  return n;
-}
-var RESERVED_OUTPUT_TOKENS;
-var init_tokenBudget = __esm({
-  "src/cli/budget/tokenBudget.ts"() {
-    "use strict";
-    init_envNumber();
-    init_historyCompaction();
-    init_observationStore();
-    init_capabilities();
-    RESERVED_OUTPUT_TOKENS = 8192;
-  }
-});
-
 // src/cli/budget/requestMeter.ts
 function estimateTokensLocal(text) {
   if (!text) return 0;
@@ -52114,6 +52260,51 @@ var init_persistCompact = __esm({
   }
 });
 
+// src/cli/budget/contextProjection.ts
+function policyFromOccupancy(occupancy, thresholds = DEFAULT_THRESHOLDS) {
+  if (occupancy >= thresholds.hardAt) return "hard";
+  if (occupancy >= thresholds.compactAt) return "compact";
+  if (occupancy >= thresholds.warnAt) return "warn";
+  return "ok";
+}
+function recordFromBudget(budget, thresholds = DEFAULT_THRESHOLDS) {
+  return {
+    occupancy: budget.occupancy,
+    estimatedHistoryTokens: budget.estimatedHistoryTokens,
+    contextLimit: budget.contextLimit,
+    ...budget.contextPressureTokens !== void 0 ? { contextPressureTokens: budget.contextPressureTokens } : {},
+    policy: policyFromOccupancy(budget.occupancy, thresholds)
+  };
+}
+function thresholdsFor(model, provider) {
+  return capabilitiesFor(model, provider).compaction;
+}
+function noteBudgetProjection(handle, record2) {
+  try {
+    handle.note("context.projection", {
+      subject: "context.projection",
+      occupancy: record2.occupancy,
+      estimatedHistoryTokens: record2.estimatedHistoryTokens,
+      contextLimit: record2.contextLimit,
+      ...record2.contextPressureTokens !== void 0 ? { contextPressureTokens: record2.contextPressureTokens } : {},
+      policy: record2.policy
+    });
+  } catch {
+  }
+}
+var DEFAULT_THRESHOLDS;
+var init_contextProjection = __esm({
+  "src/cli/budget/contextProjection.ts"() {
+    "use strict";
+    init_capabilities();
+    DEFAULT_THRESHOLDS = {
+      warnAt: 0.7,
+      compactAt: 0.85,
+      hardAt: 0.95
+    };
+  }
+});
+
 // src/cli/budget/modelContextBuilder.ts
 function messageWasRecompacted(message, history2) {
   if (message.compactedFromSeq === void 0) return false;
@@ -52220,6 +52411,15 @@ async function buildModelContext(input) {
       occupancy: Math.min(1, estimated / budget.contextLimit)
     };
   }
+  if (input.budgetNoteHandle) {
+    try {
+      noteBudgetProjection(
+        input.budgetNoteHandle,
+        recordFromBudget(budget, thresholdsFor(input.model, input.provider))
+      );
+    } catch {
+    }
+  }
   return {
     history: history2,
     requestTail,
@@ -52237,6 +52437,7 @@ var init_modelContextBuilder = __esm({
     init_tokenBudget();
     init_requestMeter();
     init_persistCompact();
+    init_contextProjection();
     init_headlessSpine();
     init_session();
   }
@@ -63135,6 +63336,20 @@ function asNumber2(v) {
 function asCallId(v, seq) {
   return typeof v === "string" && v.length > 0 ? v : `seq:${seq}`;
 }
+function parseProjection(data) {
+  const policy = asString3(data.policy);
+  return {
+    contextChars: asNumber2(data.contextChars) ?? 0,
+    returnedCount: asNumber2(data.returnedCount) ?? 0,
+    occupancy: asNumber2(data.occupancy),
+    estimatedHistoryTokens: asNumber2(data.estimatedHistoryTokens),
+    contextLimit: asNumber2(data.contextLimit),
+    contextPressureTokens: asNumber2(data.contextPressureTokens),
+    durationMs: asNumber2(data.durationMs),
+    backend: asString3(data.backend) || void 0,
+    policy: policy === "ok" || policy === "warn" || policy === "compact" || policy === "hard" ? policy : void 0
+  };
+}
 function newTurn(index, userText) {
   return {
     index,
@@ -63216,17 +63431,14 @@ function deriveHarnessState(events) {
       }
       case "session.compacted": {
         support.compactions += 1;
-        const saved = asNumber2(e.data.tokensSaved);
+        const saved = asNumber2(e.data.savedTokens) ?? asNumber2(e.data.tokensSaved);
         if (saved !== void 0) tokensSaved = (tokensSaved ?? 0) + saved;
         break;
       }
       case "note": {
         const subject = asString3(e.data.subject);
         if (subject === "context.projection") {
-          support.contextProjections.push({
-            contextChars: asNumber2(e.data.contextChars) ?? 0,
-            returnedCount: asNumber2(e.data.returnedCount) ?? 0
-          });
+          support.contextProjections.push(parseProjection(e.data));
         } else if (subject === "memory_event") {
           support.memoryEvents += 1;
         }
@@ -64177,6 +64389,7 @@ async function runOneTurn(opts, provider, model, providerStream, extras) {
     toolSpecs: typeof toolRegistry.fingerprints === "function" ? toolRegistry.fingerprints() : void 0
   });
   spineHolder.current = spine;
+  flushMemorySpineNotes(spineHolder);
   const seededHistory = await seedHeadlessModelHistory(spine, opts.history);
   emitEvent(sessionStartedEvent(spine));
   if (opts.orchestrationDecision) {
@@ -64314,6 +64527,8 @@ async function runOneTurn(opts, provider, model, providerStream, extras) {
     tools,
     sessionId: spine.sessionId,
     providerStream,
+    // T4-S2: budget occupancy/policy onto the spine (context.projection note).
+    budgetNoteHandle: spine,
     onCompactionMetric: (metrics) => recordCompactionMetrics(spine.sessionId, provider, model, metrics),
     persistCompaction: async (payload) => {
       await spine.appendEvent({
@@ -66058,6 +66273,7 @@ async function runHeadlessCouncil(opts, provider, model, providerStream, extras)
     workspace: cwd
   });
   spineHolder.current = spine;
+  flushMemorySpineNotes(spineHolder);
   const seededHistory = await seedHeadlessModelHistory(spine, opts.history);
   emitEvent(sessionStartedEvent(spine));
   if (opts.orchestrationDecision) {
@@ -66097,6 +66313,8 @@ async function runHeadlessCouncil(opts, provider, model, providerStream, extras)
     tools: contextTools,
     sessionId: spine.sessionId,
     providerStream,
+    // T4-S2: budget occupancy/policy onto the spine (context.projection note).
+    budgetNoteHandle: spine,
     onCompactionMetric: (metrics) => recordCompactionMetrics(spine.sessionId, provider, model, metrics),
     persistCompaction: async (payload) => {
       await spine.appendEvent({
@@ -66121,11 +66339,14 @@ async function runHeadlessCouncil(opts, provider, model, providerStream, extras)
     const { composeProjectContext: composeProjectContext2 } = await Promise.resolve().then(() => (init_composeContext(), composeContext_exports));
     const { loadDurableContext: loadDurableContext2 } = await Promise.resolve().then(() => (init_loadDurableContext(), loadDurableContext_exports));
     const durableState = await loadDurableContext2(cwd);
+    const { resolveRetrievalPolicy: resolveRetrievalPolicy2 } = await Promise.resolve().then(() => (init_retrievalPolicy(), retrievalPolicy_exports));
+    const retrieval = resolveRetrievalPolicy2(councilContext.budget.occupancy);
     const memoryContext = nativeMemory ? (await nativeMemory.buildContext({
       text: effectiveTask,
       useGraph: true,
-      maxChars: 2e3,
-      maxMemories: 8
+      maxChars: retrieval.maxChars,
+      maxMemories: retrieval.maxMemories,
+      ...retrieval.weights ? { weights: retrieval.weights } : {}
     })).text : "";
     const composed = composeProjectContext2({
       mode: "council",
@@ -66312,6 +66533,8 @@ async function runHeadlessZelari(opts, provider, model, providerStream, extras)
     tools: contextTools,
     sessionId: spine.sessionId,
     providerStream,
+    // T4-S2: budget occupancy/policy onto the spine (context.projection note).
+    budgetNoteHandle: spine,
     onCompactionMetric: (metrics) => recordCompactionMetrics(spine.sessionId, provider, model, metrics),
     persistCompaction: async (payload) => {
       await spine.appendEvent({
@@ -70043,6 +70266,9 @@ __export(inspectSession_exports, {
 });
 import path89 from "node:path";
 import { existsSync as existsSync56 } from "node:fs";
+function formatLimit(limit) {
+  return `${Math.round(limit / 1e3)}k`;
+}
 function renderInspectReport(state3) {
   const lines = [
     `session ${state3.session.sessionId}  status=${state3.session.status}  turns=${state3.execution.turnsTotal}`
@@ -70061,9 +70287,8 @@ function renderInspectReport(state3) {
   lines.push("support lens:");
   const projections = state3.support.contextProjections;
   const last = projections[projections.length - 1];
-  lines.push(
-    `  context projections: ${projections.length}` + (last ? `  (last: ${last.contextChars} chars \u2192 ${last.returnedCount} items)` : "")
-  );
+  const tail2 = !last ? "" : last.occupancy !== void 0 && last.contextLimit !== void 0 ? `  (last: ${Math.round(last.occupancy * 100)}% ${last.policy ?? "?"} (limit ${formatLimit(last.contextLimit)}))` : `  (last: ${last.contextChars} chars \u2192 ${last.returnedCount} items)`;
+  lines.push(`  context projections: ${projections.length}${tail2}`);
   lines.push(`  memory events: ${state3.support.memoryEvents}`);
   const saved = state3.support.tokensSavedByCompaction;
   lines.push(
@@ -73339,21 +73564,27 @@ function createPermissionAskHandler(opts) {
     const catLabel = cats.length === 1 ? cats[0] : cats.join("+") || "action";
     const title = `Allow tool "${req.toolName}"?`;
     const detail = req.reason + (cats.length ? ` [${cats.join(", ")}]` : "");
+    const claimLines = (req.claims ?? []).map((c) => `\xB7 ${c.summary}`);
+    const note = req.policyNote ? `
+${req.policyNote}` : "";
+    const claimsBlock = claimLines.length ? `
+Claims:
+${claimLines.join("\n")}` : "";
     appendSystem2?.(
       `[permission] ${title}
-${detail}
+${detail}${note}${claimsBlock}
 \u2192 Allow once \xB7 Always (tool) \xB7 Always (${catLabel}) \xB7 Deny`,
       Date.now()
     );
     let settled = false;
     const askTimeoutMs = askUserTimeoutMs();
     let cancelAskTimeout = () => void 0;
-    const finish2 = (ok, note) => {
+    const finish2 = (ok, note2) => {
       if (settled) return;
       settled = true;
       cancelAskTimeout();
       setPicker2(null);
-      if (note) appendSystem2?.(note, Date.now());
+      if (note2) appendSystem2?.(note2, Date.now());
       resolve9(ok);
     };
     cancelAskTimeout = armPickerTimeout(
@@ -74487,6 +74718,7 @@ _(${req.context})_` : "") + "\n\u2192 scegli dalla lista (il membro riprende dop
   let councilMemory;
   let councilMemoryAutoWrite = false;
   let nativeMemoryContext = "";
+  let councilRetrievalBand = "low";
   try {
     const memoryFactory = await Promise.resolve().then(() => (init_serviceFactory(), serviceFactory_exports));
     if (memoryFactory.isMemoryV2Enabled()) {
@@ -74501,11 +74733,17 @@ _(${req.context})_` : "") + "\n\u2192 scegli dalla lista (il membro riprende dop
       });
       councilMemoryAutoWrite = memoryFactory.isMemoryAutoWriteEnabled();
       if (!overrides.ragContext) {
+        const { resolveRetrievalPolicy: resolveRetrievalPolicy2 } = await Promise.resolve().then(() => (init_retrievalPolicy(), retrievalPolicy_exports));
+        const retrieval = resolveRetrievalPolicy2(
+          councilContext.budget.occupancy
+        );
+        councilRetrievalBand = retrieval.band;
         nativeMemoryContext = (await councilMemory.buildContext({
           text: effectiveText,
           useGraph: true,
-          maxChars: 2e3,
-          maxMemories: 8
+          maxChars: retrieval.maxChars,
+          maxMemories: retrieval.maxMemories,
+          ...retrieval.weights ? { weights: retrieval.weights } : {}
         })).text;
       }
     }
@@ -76772,6 +77010,7 @@ async function handleKrakenGraph(ctx, prompt) {
     onWarning: (warning) => appendSystem(ctx.setMessages, warning),
     onEvent: memorySinkFor(tuiSpineHolder)
   }) : void 0;
+  flushMemorySpineNotes(tuiSpineHolder);
   const audit = new AuditLogger();
   const taskToolDeps = {
     createSubAgentContext: createKrakenSubAgentContextFactory({