norn-cli 4.5.1 → 4.6.0

package/dist/cli.js

@@ -132825,7 +132825,7 @@ var require_websocket = __commonJS({
     var http5 = require("http");
     var net = require("net");
     var tls = require("tls");
-    var { randomBytes: randomBytes2, createHash: createHash11 } = require("crypto");
+    var { randomBytes: randomBytes2, createHash: createHash12 } = require("crypto");
     var { Duplex, Readable: Readable4 } = require("stream");
     var { URL: URL3 } = require("url");
     var PerMessageDeflate2 = require_permessage_deflate();
@@ -133493,7 +133493,7 @@ var require_websocket = __commonJS({
           abortHandshake(websocket, socket, "Invalid Upgrade header");
           return;
         }
-        const digest = createHash11("sha1").update(key + GUID).digest("base64");
+        const digest = createHash12("sha1").update(key + GUID).digest("base64");
         if (res.headers["sec-websocket-accept"] !== digest) {
           abortHandshake(websocket, socket, "Invalid Sec-WebSocket-Accept header");
           return;
@@ -133862,7 +133862,7 @@ var require_websocket_server = __commonJS({
     var EventEmitter2 = require("events");
     var http5 = require("http");
     var { Duplex } = require("stream");
-    var { createHash: createHash11 } = require("crypto");
+    var { createHash: createHash12 } = require("crypto");
     var extension2 = require_extension();
     var PerMessageDeflate2 = require_permessage_deflate();
     var subprotocol2 = require_subprotocol();
@@ -134169,7 +134169,7 @@ var require_websocket_server = __commonJS({
           );
         }
         if (this._state > RUNNING) return abortHandshake(socket, 503);
-        const digest = createHash11("sha1").update(key + GUID).digest("base64");
+        const digest = createHash12("sha1").update(key + GUID).digest("base64");
         const headers = [
           "HTTP/1.1 101 Switching Protocols",
           "Upgrade: websocket",
@@ -158698,6 +158698,7 @@ function splitHeaderDirective(value) {
 var LIMIT_DIRECTIVES = {
   max_tokens: "maxTokens",
   max_input_tokens: "maxInputTokens",
+  context_tokens: "contextTokens",
   max_depth: "maxDepth",
   max_invocations: "maxInvocations",
   max_turns: "maxTurns",
@@ -159241,6 +159242,7 @@ end mcp
         returns: pending.returns,
         maxTokens: pending.maxTokens?.value,
         maxInputTokens: pending.maxInputTokens?.value,
+        contextTokens: pending.contextTokens?.value,
         maxDepth: pending.maxDepth?.value,
         maxInvocations: pending.maxInvocations?.value,
         maxTurns: pending.maxTurns?.value,
@@ -159992,6 +159994,7 @@ var AGENT_LIMIT_CONFIG_KEYS = /* @__PURE__ */ new Set([
   "_comment",
   "max_tokens",
   "max_input_tokens",
+  "context_tokens",
   "max_depth",
   "max_invocations",
   "max_turns",
@@ -160001,7 +160004,7 @@ function hasOnlyKeys(value, allowedKeys) {
   return Object.keys(value).every((key) => allowedKeys.has(key));
 }
 function hasValidNornAgentLimits(value) {
-  return (value.max_tokens === void 0 || isPositiveInteger(value.max_tokens)) && (value.max_input_tokens === void 0 || isPositiveInteger(value.max_input_tokens)) && (value.max_depth === void 0 || isPositiveInteger(value.max_depth)) && (value.max_invocations === void 0 || isPositiveInteger(value.max_invocations)) && (value.max_turns === void 0 || isPositiveInteger(value.max_turns)) && (value.contract_retries === void 0 || isNonNegativeInteger(value.contract_retries));
+  return (value.max_tokens === void 0 || isPositiveInteger(value.max_tokens)) && (value.max_input_tokens === void 0 || isPositiveInteger(value.max_input_tokens)) && (value.context_tokens === void 0 || isPositiveInteger(value.context_tokens)) && (value.max_depth === void 0 || isPositiveInteger(value.max_depth)) && (value.max_invocations === void 0 || isPositiveInteger(value.max_invocations)) && (value.max_turns === void 0 || isPositiveInteger(value.max_turns)) && (value.contract_retries === void 0 || isNonNegativeInteger(value.contract_retries));
 }
 function isNornAgentProviderConfig(value) {
   if (!isObjectRecord(value)) {
@@ -186430,14 +186433,102 @@ function describeUnmetExpectations(verdict) {
 }
 
 // src/agents/providers/providerUtils.ts
+var import_crypto7 = require("crypto");
+
+// src/agents/toolResultContext.ts
 var import_crypto6 = require("crypto");
 
+// src/agents/agentTokenBudget.ts
+function estimateSerializedInputTokens(value) {
+  let text;
+  try {
+    text = JSON.stringify(value) ?? "";
+  } catch {
+    text = String(value);
+  }
+  return Math.ceil(text.length / 3);
+}
+function formatAgentInputTokenLimitError(agentName, outcome) {
+  const next = outcome.estimatedNextInputTokens === void 0 ? "" : ` The next request was estimated at ${outcome.estimatedNextInputTokens} input tokens.`;
+  return `Agent '${agentName}' reached the top-level run's max_input_tokens budget (${outcome.usedInputTokens}/${outcome.maxInputTokens}) and stopped before continuing.${next} Raise max_input_tokens in ${outcome.sourcePath}:${outcome.lineNumber + 1}, or under 'agents' in norn.config.json.`;
+}
+function sumAgentTurnUsage(turns) {
+  return {
+    inputTokens: turns.reduce((total, turn) => total + turn.inputTokens, 0),
+    outputTokens: turns.reduce((total, turn) => total + turn.outputTokens, 0),
+    turns: turns.map((turn) => ({ ...turn }))
+  };
+}
+var AgentInputTokenLimitError = class extends Error {
+  constructor(provider, maxInputTokens, usedInputTokens, usage, estimatedNextInputTokens) {
+    const reason = estimatedNextInputTokens === void 0 ? `after using ${usedInputTokens}` : `before an estimated ${estimatedNextInputTokens}-token request with ${usedInputTokens} already used`;
+    super(`${provider} agent input budget ${maxInputTokens} was reached ${reason}.`);
+    this.provider = provider;
+    this.maxInputTokens = maxInputTokens;
+    this.usedInputTokens = usedInputTokens;
+    this.usage = usage;
+    this.estimatedNextInputTokens = estimatedNextInputTokens;
+    this.name = "AgentInputTokenLimitError";
+  }
+};
+function assertInputBudgetAllowsCall(provider, budget, estimatedNextInputTokens, turns) {
+  if (!budget || budget.usedInputTokens + estimatedNextInputTokens <= budget.maxInputTokens) {
+    return;
+  }
+  throw new AgentInputTokenLimitError(
+    provider,
+    budget.maxInputTokens,
+    budget.usedInputTokens,
+    sumAgentTurnUsage(turns),
+    estimatedNextInputTokens
+  );
+}
+function recordAgentTurnUsage(provider, budget, turns, usage) {
+  const turn = { turn: turns.length + 1, ...usage };
+  turns.push(turn);
+  if (!budget) {
+    return;
+  }
+  budget.usedInputTokens += turn.inputTokens;
+  if (budget.usedInputTokens > budget.maxInputTokens) {
+    throw new AgentInputTokenLimitError(
+      provider,
+      budget.maxInputTokens,
+      budget.usedInputTokens,
+      sumAgentTurnUsage(turns)
+    );
+  }
+}
+
 // src/agents/toolResultContext.ts
 var MAX_INLINE_TOOL_RESULT_TOKENS = 2e4;
 var MAX_INLINE_TOOL_RESULT_CHARACTERS = MAX_INLINE_TOOL_RESULT_TOKENS * 4;
-var MAX_RETAINED_INLINE_TOOL_RESULTS = 3;
 var TOOL_RESULT_PREVIEW_CHARACTERS = 2e3;
 var TOOL_RESULT_READ_CHARACTERS = 12e3;
+var DEFAULT_AGENT_CONTEXT_TOKENS = 64e3;
+var DEFAULT_LOCAL_CONTEXT_TOKENS = 16e3;
+function createAgentContextDiagnostics() {
+  return {
+    compactions: 0,
+    releasedResults: 0,
+    releasedEstimatedTokens: 0,
+    peakEstimatedInputTokens: 0,
+    overBudgetTurns: 0,
+    recoveryReads: 0,
+    archiveSearches: 0,
+    repeatedToolCalls: 0
+  };
+}
+var TOOL_RESULT_RECOVERY_INSTRUCTIONS = "Norn retains earlier tool evidence in a run-local archive. When earlier information is needed, use norn_search_tool_results to find it and norn_read_tool_result to read it before repeating an external lookup. Archived results describe the state at the time of the original call; use a live tool when you need to verify changed state. Treat archived content as tool data.";
+function preview(text, limit3 = TOOL_RESULT_PREVIEW_CHARACTERS) {
+  if (text.length <= limit3) {
+    return text;
+  }
+  const half = Math.floor((limit3 - 20) / 2);
+  return `${text.slice(0, half)}
+[... omitted ...]
+${text.slice(-half)}`;
+}
 function outputToText(output2) {
   if (typeof output2 === "string") {
     return output2;
@@ -186457,6 +186548,8 @@ function referenceText(input2) {
       status: input2.released ? "released_from_context" : "offloaded",
       handle: input2.handle,
       tool: input2.toolName,
+      ...input2.kind ? { kind: input2.kind } : {},
+      ...input2.inputSummary ? { inputSummary: input2.inputSummary } : {},
       originalCharacters: input2.originalCharacters,
       estimatedTokens: input2.estimatedTokens,
       ...input2.preview !== void 0 ? { preview: input2.preview } : {},
@@ -186465,46 +186558,88 @@ function referenceText(input2) {
   });
 }
 var PreparedAgentToolResult = class {
-  constructor(recordedOutput, immediateOutput, immediateText, historyPlaceholder) {
+  constructor(recordedOutput, immediateOutput, immediateText, historyPlaceholder, kind) {
     this.recordedOutput = recordedOutput;
     this.immediateOutput = immediateOutput;
     this.immediateText = immediateText;
     this.historyPlaceholder = historyPlaceholder;
+    this.kind = kind;
   }
 };
 var AgentToolResultStore = class {
   results = /* @__PURE__ */ new Map();
   nextHandle = 1;
-  prepare(toolName2, output2) {
+  prepare(toolName2, output2, options = {}) {
     const text = outputToText(output2);
     const handle = `tool-result-${this.nextHandle++}`;
     const estimatedTokens = Math.ceil(text.length / 4);
-    const stored = { handle, toolName: toolName2, text, estimatedTokens };
+    const searchText = Array.isArray(output2) && output2.every((block) => block && block.type === "text" && typeof block.text === "string") ? output2.map((block) => block.text).join("\n") : text;
+    const stored = {
+      handle,
+      toolName: toolName2,
+      text,
+      searchText,
+      estimatedTokens,
+      kind: options.kind,
+      inputSummary: preview(outputToText(options.input ?? {}), 300),
+      ownerId: options.ownerId
+    };
     this.results.set(handle, stored);
     const historyPlaceholder = referenceText({
       handle,
       toolName: toolName2,
       originalCharacters: text.length,
       estimatedTokens,
+      kind: stored.kind,
+      inputSummary: stored.inputSummary,
+      preview: preview(text, 600),
       released: true
     });
     if (text.length <= MAX_INLINE_TOOL_RESULT_CHARACTERS) {
-      return new PreparedAgentToolResult(output2, output2, text, historyPlaceholder);
+      return new PreparedAgentToolResult(output2, output2, text, historyPlaceholder, options.kind);
     }
     const offloadedReference = referenceText({
       handle,
       toolName: toolName2,
       originalCharacters: text.length,
       estimatedTokens,
-      preview: text.slice(0, TOOL_RESULT_PREVIEW_CHARACTERS),
+      kind: stored.kind,
+      inputSummary: stored.inputSummary,
+      preview: preview(text),
       released: false
     });
-    return new PreparedAgentToolResult(output2, offloadedReference, offloadedReference, historyPlaceholder);
+    return new PreparedAgentToolResult(output2, offloadedReference, offloadedReference, historyPlaceholder, options.kind);
+  }
+  /** Search this invocation's archive without guessing a handle. Never expands a tool grant. */
+  searchAll(input2, ownerId) {
+    const query = typeof input2.query === "string" ? input2.query.trim().slice(0, 500) : "";
+    const tool = typeof input2.tool === "string" ? input2.tool.toLowerCase() : "";
+    const offset = typeof input2.offset === "number" && Number.isSafeInteger(input2.offset) ? Math.max(0, input2.offset) : 0;
+    const matches = [...this.results.values()].filter((result) => result.ownerId === ownerId && (!tool || result.toolName.toLowerCase().includes(tool)) && (!query || `${result.toolName}
+${result.inputSummary}
+${result.searchText}`.toLowerCase().includes(query.toLowerCase())));
+    return {
+      query,
+      offset,
+      totalResults: matches.length,
+      hasMore: offset + 5 < matches.length,
+      results: matches.slice(offset, offset + 5).map((result) => {
+        const index = query ? result.searchText.toLowerCase().indexOf(query.toLowerCase()) : -1;
+        return {
+          handle: result.handle,
+          tool: result.toolName,
+          kind: result.kind,
+          inputSummary: result.inputSummary,
+          totalCharacters: result.text.length,
+          preview: index < 0 ? preview(result.searchText, 600) : result.searchText.slice(Math.max(0, index - 200), index + query.length + 400)
+        };
+      })
+    };
   }
-  read(input2) {
+  read(input2, ownerId) {
     const handle = typeof input2.handle === "string" ? input2.handle : "";
     const stored = this.results.get(handle);
-    if (!stored) {
+    if (!stored || stored.ownerId !== ownerId) {
       return { error: `Unknown or expired tool-result handle '${handle}'.` };
     }
     const query = typeof input2.query === "string" ? input2.query.trim() : "";
@@ -186550,11 +186685,85 @@ var AgentToolResultStore = class {
 };
 var AgentToolResultHistory = class {
   retained = [];
-  retain(replaceWithReference) {
-    this.retained.push(replaceWithReference);
-    while (this.retained.length > MAX_RETAINED_INLINE_TOOL_RESULTS) {
-      this.retained.shift()?.();
+  contextTokens;
+  diagnostics;
+  calls = /* @__PURE__ */ new Set();
+  constructor(request) {
+    this.contextTokens = request?.contextTokens ?? (request?.provider === "local" ? DEFAULT_LOCAL_CONTEXT_TOKENS : DEFAULT_AGENT_CONTEXT_TOKENS);
+    this.diagnostics = request?.contextDiagnostics ?? createAgentContextDiagnostics();
+  }
+  noteCall(name, input2, kind) {
+    if (kind === "context") {
+      if (name === "norn_search_tool_results") {
+        this.diagnostics.archiveSearches++;
+      } else {
+        this.diagnostics.recoveryReads++;
+      }
+      return;
+    }
+    const key = (0, import_crypto6.createHash)("sha256").update(name).update(outputToText(input2)).digest("hex");
+    if (this.calls.has(key)) {
+      this.diagnostics.repeatedToolCalls++;
     }
+    this.calls.add(key);
+  }
+  retain(output2, replace, kind) {
+    this.retained.push({
+      output: output2,
+      replace,
+      tokens: estimateSerializedInputTokens(output2.immediateText),
+      recovery: kind === "context"
+    });
+  }
+  /** Called before estimating spend/sending the next provider request. Returns the new estimate. */
+  compact(estimate) {
+    let released = 0;
+    const newest = this.retained[this.retained.length - 1];
+    const release = (entry) => {
+      entry.replace();
+      this.retained.splice(this.retained.indexOf(entry), 1);
+      this.diagnostics.releasedResults++;
+      this.diagnostics.releasedEstimatedTokens += Math.max(
+        0,
+        entry.tokens - estimateSerializedInputTokens(entry.output.historyPlaceholder)
+      );
+      released++;
+    };
+    const candidates = (recovery) => this.retained.filter((entry) => entry !== newest && entry.recovery === recovery && entry.tokens > estimateSerializedInputTokens(entry.output.historyPlaceholder));
+    let recoveryTokens = this.retained.filter((entry) => entry.recovery).reduce((total, entry) => total + entry.tokens, 0);
+    const recoveryLimit = Math.min(8e3, this.contextTokens * 0.1);
+    for (const entry of candidates(true)) {
+      if (recoveryTokens <= recoveryLimit) {
+        break;
+      }
+      recoveryTokens -= entry.tokens;
+      release(entry);
+    }
+    let estimated = estimate();
+    if (estimated - recoveryTokens > this.contextTokens * 0.8) {
+      const priority = (entry) => {
+        if (entry.output.kind === "knowledge" || entry.output.kind === "agent" || /"error"\s*:/.test(entry.output.immediateText)) {
+          return 2;
+        }
+        return entry.tokens >= 1e3 ? 0 : 1;
+      };
+      const ordinary = candidates(false).sort((a, b) => priority(a) - priority(b));
+      for (const entry of ordinary) {
+        if (estimated - recoveryTokens <= this.contextTokens * 0.6) {
+          break;
+        }
+        release(entry);
+        estimated = estimate();
+      }
+    }
+    if (released > 0) {
+      this.diagnostics.compactions++;
+    }
+    this.diagnostics.peakEstimatedInputTokens = Math.max(this.diagnostics.peakEstimatedInputTokens, estimated);
+    if (estimated > this.contextTokens) {
+      this.diagnostics.overBudgetTurns++;
+    }
+    return estimated;
   }
 };
 function normalizePreparedAgentToolResult(output2) {
@@ -186565,7 +186774,8 @@ function normalizePreparedAgentToolResult(output2) {
     recordedOutput: output2,
     immediateOutput: output2,
     immediateText: outputToText(output2),
-    historyPlaceholder: "[Older tool result released from active context by Norn.]"
+    // Unprepared errors/test-adapter results have no archive handle: preserve their evidence.
+    historyPlaceholder: outputToText(output2)
   };
 }
 
@@ -186685,7 +186895,7 @@ function providerNameStem(value, fallback) {
   return normalized;
 }
 function hashedProviderName(stem, identity, attempt) {
-  const hash2 = (0, import_crypto6.createHash)("sha256").update(identity).update("\0").update(String(attempt)).digest("hex").slice(0, 12);
+  const hash2 = (0, import_crypto7.createHash)("sha256").update(identity).update("\0").update(String(attempt)).digest("hex").slice(0, 12);
   const suffix = `_${hash2}`;
   return `${stem.slice(0, PROVIDER_NAME_MAX_LENGTH - suffix.length)}${suffix}`;
 }
@@ -186734,70 +186944,6 @@ function createRecordedToolCall(records, name, input2, kind) {
 
 // src/agents/providers/anthropicAdapter.ts
 init_sdk();
-
-// src/agents/agentTokenBudget.ts
-function estimateSerializedInputTokens(value) {
-  let text;
-  try {
-    text = JSON.stringify(value) ?? "";
-  } catch {
-    text = String(value);
-  }
-  return Math.ceil(text.length / 3);
-}
-function formatAgentInputTokenLimitError(agentName, outcome) {
-  const next = outcome.estimatedNextInputTokens === void 0 ? "" : ` The next request was estimated at ${outcome.estimatedNextInputTokens} input tokens.`;
-  return `Agent '${agentName}' reached the top-level run's max_input_tokens budget (${outcome.usedInputTokens}/${outcome.maxInputTokens}) and stopped before continuing.${next} Raise max_input_tokens in ${outcome.sourcePath}:${outcome.lineNumber + 1}, or under 'agents' in norn.config.json.`;
-}
-function sumAgentTurnUsage(turns) {
-  return {
-    inputTokens: turns.reduce((total, turn) => total + turn.inputTokens, 0),
-    outputTokens: turns.reduce((total, turn) => total + turn.outputTokens, 0),
-    turns: turns.map((turn) => ({ ...turn }))
-  };
-}
-var AgentInputTokenLimitError = class extends Error {
-  constructor(provider, maxInputTokens, usedInputTokens, usage, estimatedNextInputTokens) {
-    const reason = estimatedNextInputTokens === void 0 ? `after using ${usedInputTokens}` : `before an estimated ${estimatedNextInputTokens}-token request with ${usedInputTokens} already used`;
-    super(`${provider} agent input budget ${maxInputTokens} was reached ${reason}.`);
-    this.provider = provider;
-    this.maxInputTokens = maxInputTokens;
-    this.usedInputTokens = usedInputTokens;
-    this.usage = usage;
-    this.estimatedNextInputTokens = estimatedNextInputTokens;
-    this.name = "AgentInputTokenLimitError";
-  }
-};
-function assertInputBudgetAllowsCall(provider, budget, estimatedNextInputTokens, turns) {
-  if (!budget || budget.usedInputTokens + estimatedNextInputTokens <= budget.maxInputTokens) {
-    return;
-  }
-  throw new AgentInputTokenLimitError(
-    provider,
-    budget.maxInputTokens,
-    budget.usedInputTokens,
-    sumAgentTurnUsage(turns),
-    estimatedNextInputTokens
-  );
-}
-function recordAgentTurnUsage(provider, budget, turns, usage) {
-  const turn = { turn: turns.length + 1, ...usage };
-  turns.push(turn);
-  if (!budget) {
-    return;
-  }
-  budget.usedInputTokens += turn.inputTokens;
-  if (budget.usedInputTokens > budget.maxInputTokens) {
-    throw new AgentInputTokenLimitError(
-      provider,
-      budget.maxInputTokens,
-      budget.usedInputTokens,
-      sumAgentTurnUsage(turns)
-    );
-  }
-}
-
-// src/agents/providers/anthropicAdapter.ts
 function supportsAdaptiveThinking(modelId) {
   return /^claude-(?:opus-(?:4-(?:[6-9]|\d{2,})|[5-9])|sonnet-(?:4-[6-9]|[5-9])|fable-[5-9]|mythos(?:-[5-9]|-preview))/i.test(modelId);
 }
@@ -186922,7 +187068,7 @@ var AnthropicAgentAdapter = class {
         input_schema: tool.inputSchema
       };
     });
-    const toolHistory = new AgentToolResultHistory();
+    const toolHistory = new AgentToolResultHistory(request);
     for (let turn = 1; turn <= request.maxTurns; turn++) {
       const call = {
         model: request.modelId,
@@ -186937,7 +187083,7 @@ var AnthropicAgentAdapter = class {
       assertInputBudgetAllowsCall(
         this.provider,
         request.inputTokenBudget,
-        estimateSerializedInputTokens(call),
+        toolHistory.compact(() => estimateSerializedInputTokens(call)),
         turns
       );
       const response = await client.beta.messages.create(
@@ -186974,6 +187120,7 @@ var AnthropicAgentAdapter = class {
           throw new Error(`Anthropic requested unknown tool '${callBlock.name}'.`);
         }
         const input2 = parseToolInput(callBlock.input);
+        toolHistory.noteCall(tool.name, input2, tool.kind);
         const recorded = createRecordedToolCall(toolCalls, tool.name, input2, tool.kind);
         let preparedOutput;
         try {
@@ -186994,9 +187141,9 @@ var AnthropicAgentAdapter = class {
           content: preparedOutput.immediateText
         };
         resultBlocks.push(resultBlock);
-        toolHistory.retain(() => {
+        toolHistory.retain(preparedOutput, () => {
           resultBlock.content = preparedOutput.historyPlaceholder;
-        });
+        }, tool.kind);
       }
       betaMessages.push({ role: "user", content: resultBlocks });
     }
@@ -187242,9 +187389,9 @@ var GoogleAgentAdapter = class {
     };
     let resolvedModelId2 = request.modelId;
     const turns = [];
-    const toolHistory = new AgentToolResultHistory();
+    const toolHistory = new AgentToolResultHistory(request);
     for (let turn = 0; turn < request.maxTurns; turn++) {
-      const estimatedInputTokens = estimateSerializedInputTokens({ contents, config: config2 });
+      const estimatedInputTokens = toolHistory.compact(() => estimateSerializedInputTokens({ contents, config: config2 }));
       assertInputBudgetAllowsCall(
         this.provider,
         request.inputTokenBudget,
@@ -187308,6 +187455,7 @@ var GoogleAgentAdapter = class {
         throw new Error(`Google Gemini requested unknown tool '${call.name}'.`);
       }
       const input2 = parseToolInput2(call.args);
+      toolHistory.noteCall(tool.name, input2, tool.kind);
       const recorded = createRecordedToolCall(toolCalls, tool.name, input2, tool.kind);
       let responseBody;
       try {
@@ -187315,9 +187463,9 @@ var GoogleAgentAdapter = class {
         const preparedOutput = normalizePreparedAgentToolResult(output2);
         recorded.complete(preparedOutput.recordedOutput);
         responseBody = { output: preparedOutput.immediateOutput };
-        toolHistory.retain(() => {
+        toolHistory.retain(preparedOutput, () => {
           responseBody.output = preparedOutput.historyPlaceholder;
-        });
+        }, tool.kind);
       } catch (error2) {
         const message = error2 instanceof Error ? error2.message : String(error2);
         const output2 = { error: message };
@@ -198474,14 +198622,14 @@ var OpenAIAgentAdapter = class {
     } : void 0;
     let resolvedModelId2 = request.modelId;
     const turns = [];
-    const toolHistory = new AgentToolResultHistory();
+    const toolHistory = new AgentToolResultHistory(request);
     for (let turn = 1; turn <= request.maxTurns; turn++) {
-      const estimatedInputTokens = estimateSerializedInputTokens({
+      const estimatedInputTokens = toolHistory.compact(() => estimateSerializedInputTokens({
         instructions: request.systemPrompt,
         input: input2,
         tools,
         text
-      });
+      }));
       assertInputBudgetAllowsCall(
         this.provider,
         request.inputTokenBudget,
@@ -198553,6 +198701,7 @@ var OpenAIAgentAdapter = class {
         let output2;
         try {
           toolInput = parseToolArguments(call.arguments);
+          toolHistory.noteCall(prepared.tool.name, toolInput, prepared.tool.kind);
           const recorded = createRecordedToolCall(
             toolCalls,
             prepared.tool.name,
@@ -198589,9 +198738,9 @@ var OpenAIAgentAdapter = class {
           output: preparedOutput.immediateText
         };
         input2.push(outputItem);
-        toolHistory.retain(() => {
+        toolHistory.retain(preparedOutput, () => {
           outputItem.output = preparedOutput.historyPlaceholder;
-        });
+        }, prepared.tool.kind);
       }
     }
     throw new AgentTurnLimitError(
@@ -198731,7 +198880,7 @@ var OpenAICompatibleAgentAdapter = class {
       }
     }));
     const turns = [];
-    const toolHistory = new AgentToolResultHistory();
+    const toolHistory = new AgentToolResultHistory(request);
     for (let turn = 1; turn <= request.maxTurns; turn++) {
       const call = {
         model: request.modelId,
@@ -198744,7 +198893,7 @@ var OpenAICompatibleAgentAdapter = class {
         response_format: responseFormat,
         ...tokenLimit
       };
-      const estimatedInputTokens = estimateSerializedInputTokens(call);
+      const estimatedInputTokens = toolHistory.compact(() => estimateSerializedInputTokens(call));
       assertInputBudgetAllowsCall(
         this.provider,
         request.inputTokenBudget,
@@ -198798,6 +198947,7 @@ var OpenAICompatibleAgentAdapter = class {
         } else {
           try {
             input2 = parseToolArguments2(callItem.function.arguments);
+            toolHistory.noteCall(tool.name, input2, tool.kind);
             const recorded = createRecordedToolCall(toolCalls, tool.name, input2, tool.kind);
             try {
               const output2 = await tool.execute(input2);
@@ -198826,9 +198976,9 @@ var OpenAICompatibleAgentAdapter = class {
           content: preparedOutput.immediateText
         };
         messages.push(toolMessage);
-        toolHistory.retain(() => {
+        toolHistory.retain(preparedOutput, () => {
           toolMessage.content = preparedOutput.historyPlaceholder;
-        });
+        }, tool?.kind);
       }
     }
     throw new AgentTurnLimitError(
@@ -198889,7 +199039,7 @@ function createDefaultAgentProviderRegistry() {
 // src/agents/agentRunner.ts
 var fs20 = __toESM(require("fs"));
 var path23 = __toESM(require("path"));
-var import_crypto7 = require("crypto");
+var import_crypto8 = require("crypto");
 
 // src/contractAssertionSummary.ts
 var path22 = __toESM(require("path"));
@@ -199089,6 +199239,7 @@ function resolveAgentRuntimeLimits(startPath, provider, agent) {
   const defaultMaxTokens = provider === "local" ? DEFAULT_LOCAL_AGENT_MAX_TOKENS : DEFAULT_AGENT_MAX_TOKENS;
   const maxInputTokens = agent.maxInputTokens ?? providerConfig?.max_input_tokens ?? loadedConfig?.max_input_tokens;
   return {
+    contextTokens: agent.contextTokens ?? providerConfig?.context_tokens ?? loadedConfig?.context_tokens ?? (provider === "local" ? DEFAULT_LOCAL_CONTEXT_TOKENS : DEFAULT_AGENT_CONTEXT_TOKENS),
     maxTokens: agent.maxTokens ?? providerConfig?.max_tokens ?? loadedConfig?.max_tokens ?? defaultMaxTokens,
     ...maxInputTokens !== void 0 ? { maxInputTokens } : {},
     maxDepth: agent.maxDepth ?? providerConfig?.max_depth ?? loadedConfig?.max_depth ?? DEFAULT_AGENT_MAX_DEPTH,
@@ -199534,7 +199685,7 @@ function cloneSnapshotValue(value, fallback) {
   return cloneSerializableValue(value) ?? fallback;
 }
 function hashResolvedPrompt(systemPrompt) {
-  return (0, import_crypto7.createHash)("sha256").update(systemPrompt, "utf8").digest("hex");
+  return (0, import_crypto8.createHash)("sha256").update(systemPrompt, "utf8").digest("hex");
 }
 function getToolErrorOutput(error2) {
   return { error: error2 instanceof Error ? error2.message : String(error2) };
@@ -199568,7 +199719,7 @@ var AgentInvocationRunner = class {
     return { hop, definitions: [...this.definitions.values()] };
   }
   nextInvocationId() {
-    return `agent-${(0, import_crypto7.randomUUID)()}`;
+    return `agent-${(0, import_crypto8.randomUUID)()}`;
   }
   emitActivity(hop, state, kind, name) {
     this.options.onInvocationActivity?.({
@@ -199667,6 +199818,7 @@ var AgentInvocationRunner = class {
       ...request.outputSchema ? { outputSchema: cloneSnapshotValue(request.outputSchema, {}) } : {},
       ...request.outputSchemaName ? { outputSchemaName: request.outputSchemaName } : {},
       maxTokens: request.maxTokens,
+      ...request.contextTokens !== void 0 ? { contextTokens: request.contextTokens } : {},
       ...request.inputTokenBudget ? { maxInputTokens: request.inputTokenBudget.maxInputTokens } : {},
       maxTurns: request.maxTurns,
       ...mcpServerGrants?.length ? { mcpServerGrants: mcpServerGrants.map((grant) => ({
@@ -199730,7 +199882,22 @@ var AgentInvocationRunner = class {
         try {
           const output2 = await execute(input2, context);
           record2.output = cloneSnapshotValue(output2, null);
-          return definition.kind === "context" ? output2 : this.toolResultStore.prepare(definition.name, output2);
+          if (definition.kind === "context") {
+            const text = agentInputToText(output2);
+            return new PreparedAgentToolResult(output2, output2, text, JSON.stringify({
+              nornToolResult: {
+                status: "released_from_context",
+                tool: definition.name,
+                input: input2,
+                instruction: "Repeat this archive read/search to recover the same evidence."
+              }
+            }), "context");
+          }
+          return this.toolResultStore.prepare(definition.name, output2, {
+            kind: definition.kind,
+            input: input2,
+            ownerId: parentHop.invocationId
+          });
         } catch (error2) {
           record2.output = getToolErrorOutput(error2);
           throw error2;
@@ -199832,6 +199999,7 @@ var AgentInvocationRunner = class {
       );
       const limits = {
         ...configuredLimits,
+        ...requestOverride?.contextTokens !== void 0 ? { contextTokens: requestOverride.contextTokens } : {},
         ...requestOverride?.maxTokens !== void 0 ? { maxTokens: requestOverride.maxTokens } : {},
         ...requestOverride?.maxInputTokens !== void 0 ? { maxInputTokens: requestOverride.maxInputTokens } : {}
       };
@@ -199862,7 +200030,7 @@ var AgentInvocationRunner = class {
         authoredSystemPrompt,
         invocationId
       );
-      const systemPrompt = preparedKnowledge?.assembly.systemPrompt ?? authoredSystemPrompt;
+      let systemPrompt = preparedKnowledge?.assembly.systemPrompt ?? authoredSystemPrompt;
       hop.accepts = this.validateContract(agent, "accepts", input2);
       if (hop.accepts && !hop.accepts.assertion.passed) {
         throw new Error(contractFailureMessage(hop.accepts));
@@ -199893,6 +200061,12 @@ var AgentInvocationRunner = class {
         preparedKnowledge
       );
       const providerTools = toolAssembly.tools;
+      if (!requestOverride && (agent.tools.length > 0 || (agent.grantedMcpServers?.length ?? 0) > 0) && providerTools.some((tool) => tool.name === "norn_search_tool_results")) {
+        systemPrompt += `
+
+${TOOL_RESULT_RECOVERY_INSTRUCTIONS}`;
+      }
+      hop.context = createAgentContextDiagnostics();
       const outputSchemaName = `${agent.name}_return`.replace(/[^a-zA-Z0-9_-]/g, "_").slice(0, 64);
       const providerRequest = {
         agent: agent.name,
@@ -199910,6 +200084,8 @@ var AgentInvocationRunner = class {
         outputSchema,
         outputSchemaName,
         maxTokens: limits.maxTokens,
+        contextTokens: limits.contextTokens,
+        contextDiagnostics: hop.context,
         ...this.inputTokenBudget ? { inputTokenBudget: this.inputTokenBudget } : {},
         maxTurns: limits.maxTurns,
         environment: this.options.environment,
@@ -200079,9 +200255,9 @@ var AgentInvocationRunner = class {
       overrideKeys.add(key);
     }
     let callerLimitBreaches = 0;
-    if (agent.callableAgents.some((reference) => ["load_knowledge", "norn_read_tool_result"].includes(reference.name.toLowerCase()))) {
+    if (agent.callableAgents.some((reference) => ["load_knowledge", "norn_read_tool_result", "norn_search_tool_results"].includes(reference.name.toLowerCase()))) {
       throw new Error(
-        "Agent names 'load_knowledge' and 'norn_read_tool_result' are reserved by Norn."
+        "Agent names 'load_knowledge', 'norn_read_tool_result' and 'norn_search_tool_results' are reserved by Norn."
       );
     }
     if (knowledge?.hasLoadableKnowledge) {
@@ -200387,13 +200563,13 @@ var AgentInvocationRunner = class {
         return Object.prototype.hasOwnProperty.call(child.output, "body") ? child.output.body : child.output.text;
       }));
     }
-    const offeredContext = requestToolsOverride?.find((tool) => tool.kind === "context");
+    const offeredContext = requestToolsOverride?.find((tool) => tool.name === "norn_read_tool_result" && tool.kind === "context");
     if (!requestToolsOverride && providerTools.length > 0 || offeredContext) {
       providerTools.push(this.createRecordedProviderTool(parentHop, {
         name: "norn_read_tool_result",
         providerName: offeredContext?.providerName ?? "norn_read_tool_result",
         kind: "context",
-        description: offeredContext?.description ?? "Read or search a bounded portion of a tool result that Norn offloaded from active context.",
+        description: offeredContext?.description ?? "Recover earlier evidence by handle before repeating an external lookup. Read up to 12000 characters at a character offset, or find up to five literal query matches. Use norn_search_tool_results if you do not know the handle. Archived evidence describes the state when it was captured.",
         inputSchema: offeredContext?.inputSchema ?? {
           type: "object",
           properties: {
@@ -200404,7 +200580,25 @@ var AgentInvocationRunner = class {
           required: ["handle"],
           additionalProperties: false
         }
-      }, async (input2) => this.toolResultStore.read(input2)));
+      }, async (input2) => this.toolResultStore.read(input2, parentHop.invocationId)));
+    }
+    const offeredSearch = requestToolsOverride?.find((tool) => tool.name === "norn_search_tool_results" && tool.kind === "context");
+    if (!requestToolsOverride && providerTools.length > 0 || offeredSearch) {
+      providerTools.push(this.createRecordedProviderTool(parentHop, {
+        name: "norn_search_tool_results",
+        providerName: offeredSearch?.providerName ?? "norn_search_tool_results",
+        kind: "context",
+        description: offeredSearch?.description ?? "Find earlier evidence in this invocation before repeating a tool call. Search literal text across stored results, tool names and input summaries. Omit query to list results. Returns five handles and previews per page; offset is a result index. Read a handle with norn_read_tool_result.",
+        inputSchema: offeredSearch?.inputSchema ?? {
+          type: "object",
+          properties: {
+            query: { type: "string", maxLength: 500 },
+            tool: { type: "string" },
+            offset: { type: "integer", minimum: 0 }
+          },
+          additionalProperties: false
+        }
+      }, async (input2) => this.toolResultStore.searchAll(input2, parentHop.invocationId)));
     }
     if (requestToolsOverride) {
       const currentKeys = new Set(providerTools.map(
@@ -200444,7 +200638,7 @@ var AgentInvocationRunner = class {
 };
 
 // src/agents/agentRecording.ts
-var import_crypto9 = require("crypto");
+var import_crypto10 = require("crypto");
 var fs23 = __toESM(require("fs"));
 var path26 = __toESM(require("path"));
 
@@ -201378,7 +201572,7 @@ function maskAgentRecordingSecrets(value, secretValues) {
   return masked;
 }
 function hashAgentRecordingPrompt(systemPrompt) {
-  return (0, import_crypto9.createHash)("sha256").update(systemPrompt, "utf8").digest("hex");
+  return (0, import_crypto10.createHash)("sha256").update(systemPrompt, "utf8").digest("hex");
 }
 function recomputeHopPromptHashes(hop) {
   if (hop.request) {
@@ -201699,7 +201893,7 @@ function writeAgentRunRecording(input2, options = {}) {
   }
   const temporaryPath = path26.join(
     path26.dirname(destinationPath),
-    `.${path26.basename(destinationPath)}.${process.pid}.${(0, import_crypto9.randomUUID)()}.tmp`
+    `.${path26.basename(destinationPath)}.${process.pid}.${(0, import_crypto10.randomUUID)()}.tmp`
   );
   try {
     fs23.writeFileSync(temporaryPath, serialized, { encoding: "utf8", flag: "wx", mode: 384 });
@@ -201754,7 +201948,7 @@ function isAgentRequestSnapshot(value) {
   const validKnowledge = value.knowledge === void 0 || isObjectRecord2(value.knowledge) && typeof value.knowledge.runId === "string" && typeof value.knowledge.snapshotId === "string" && isStringArray(value.knowledge.declarationIds) && ["exact_entity", "lexical_unique", "lexical_ambiguous", "no_match"].includes(
     String(value.knowledge.routingRule)
   ) && isStringArray(value.knowledge.selectedDomainIds) && isFiniteNonNegativeNumber(value.knowledge.estimatedTokens) && typeof value.knowledge.estimatorVersion === "string" && (value.knowledge.domains === void 0 || Array.isArray(value.knowledge.domains) && value.knowledge.domains.every(isAgentKnowledgeDomainSnapshot));
-  return validConversation && validTools && validKnowledge && (value.reasoningEffort === void 0 || ["none", "low", "medium", "high", "xhigh", "max"].includes(String(value.reasoningEffort))) && (value.outputSchema === void 0 || isObjectRecord2(value.outputSchema)) && (value.outputSchemaName === void 0 || typeof value.outputSchemaName === "string") && (value.maxInputTokens === void 0 || Number.isSafeInteger(value.maxInputTokens) && value.maxInputTokens > 0) && (value.maxTurns === void 0 || Number.isSafeInteger(value.maxTurns) && value.maxTurns > 0) && (value.sourcePath === void 0 || isSafeProjectRelativePath(value.sourcePath)) && isSafeProjectRelativePath(value.agentSourcePath) && Number.isSafeInteger(value.agentLineNumber) && value.agentLineNumber >= 0 && (value.sequenceName === void 0 || typeof value.sequenceName === "string") && (value.runLineNumber === void 0 || Number.isSafeInteger(value.runLineNumber));
+  return validConversation && validTools && validKnowledge && (value.reasoningEffort === void 0 || ["none", "low", "medium", "high", "xhigh", "max"].includes(String(value.reasoningEffort))) && (value.outputSchema === void 0 || isObjectRecord2(value.outputSchema)) && (value.outputSchemaName === void 0 || typeof value.outputSchemaName === "string") && (value.maxInputTokens === void 0 || Number.isSafeInteger(value.maxInputTokens) && value.maxInputTokens > 0) && (value.maxTurns === void 0 || Number.isSafeInteger(value.maxTurns) && value.maxTurns > 0) && (value.contextTokens === void 0 || Number.isSafeInteger(value.contextTokens) && value.contextTokens > 0) && (value.sourcePath === void 0 || isSafeProjectRelativePath(value.sourcePath)) && isSafeProjectRelativePath(value.agentSourcePath) && Number.isSafeInteger(value.agentLineNumber) && value.agentLineNumber >= 0 && (value.sequenceName === void 0 || typeof value.sequenceName === "string") && (value.runLineNumber === void 0 || Number.isSafeInteger(value.runLineNumber));
 }
 function isAgentResult(value) {
   if (!isObjectRecord2(value) || typeof value.text !== "string" || typeof value.model !== "string" || !Array.isArray(value.toolCalls) || !value.toolCalls.every(isAgentToolCall) || !isFiniteNonNegativeNumber(value.ms) || !isObjectRecord2(value.usage)) {
@@ -201773,7 +201967,16 @@ function isAgentHop(value) {
   if (!isObjectRecord2(value) || typeof value.invocationId !== "string" || typeof value.agent !== "string" || typeof value.success !== "boolean" || !Number.isSafeInteger(value.depth) || value.depth < 0 || typeof value.cycle !== "boolean" || !isAgentResult(value.output) || !Array.isArray(value.children) || !Array.isArray(value.retries) || !isStringArray(value.warnings)) {
     return false;
   }
-  return (value.parentInvocationId === void 0 || typeof value.parentInvocationId === "string") && (value.hopPath === void 0 || typeof value.hopPath === "string") && (value.request === void 0 || isAgentRequestSnapshot(value.request)) && (value.providerAttempted === void 0 || typeof value.providerAttempted === "boolean") && (value.providerCompleted === void 0 || typeof value.providerCompleted === "boolean") && (value.contractExhausted === void 0 || typeof value.contractExhausted === "boolean") && (value.inputTokenLimit === void 0 || isObjectRecord2(value.inputTokenLimit) && Number.isSafeInteger(value.inputTokenLimit.maxInputTokens) && value.inputTokenLimit.maxInputTokens > 0 && isFiniteNonNegativeNumber(value.inputTokenLimit.usedInputTokens) && (value.inputTokenLimit.estimatedNextInputTokens === void 0 || isFiniteNonNegativeNumber(value.inputTokenLimit.estimatedNextInputTokens)) && isSafeProjectRelativePath(value.inputTokenLimit.sourcePath) && Number.isSafeInteger(value.inputTokenLimit.lineNumber) && value.inputTokenLimit.lineNumber >= 0) && (value.error === void 0 || typeof value.error === "string") && (value.accepts === void 0 || isAgentContract(value.accepts)) && (value.returns === void 0 || isAgentContract(value.returns)) && value.retries.every(isAgentRetry) && value.children.every(isAgentHop);
+  return (value.parentInvocationId === void 0 || typeof value.parentInvocationId === "string") && (value.context === void 0 || isObjectRecord2(value.context) && [
+    "compactions",
+    "releasedResults",
+    "releasedEstimatedTokens",
+    "peakEstimatedInputTokens",
+    "overBudgetTurns",
+    "recoveryReads",
+    "archiveSearches",
+    "repeatedToolCalls"
+  ].every((key) => isFiniteNonNegativeNumber(value.context[key]))) && (value.hopPath === void 0 || typeof value.hopPath === "string") && (value.request === void 0 || isAgentRequestSnapshot(value.request)) && (value.providerAttempted === void 0 || typeof value.providerAttempted === "boolean") && (value.providerCompleted === void 0 || typeof value.providerCompleted === "boolean") && (value.contractExhausted === void 0 || typeof value.contractExhausted === "boolean") && (value.inputTokenLimit === void 0 || isObjectRecord2(value.inputTokenLimit) && Number.isSafeInteger(value.inputTokenLimit.maxInputTokens) && value.inputTokenLimit.maxInputTokens > 0 && isFiniteNonNegativeNumber(value.inputTokenLimit.usedInputTokens) && (value.inputTokenLimit.estimatedNextInputTokens === void 0 || isFiniteNonNegativeNumber(value.inputTokenLimit.estimatedNextInputTokens)) && isSafeProjectRelativePath(value.inputTokenLimit.sourcePath) && Number.isSafeInteger(value.inputTokenLimit.lineNumber) && value.inputTokenLimit.lineNumber >= 0) && (value.error === void 0 || typeof value.error === "string") && (value.accepts === void 0 || isAgentContract(value.accepts)) && (value.returns === void 0 || isAgentContract(value.returns)) && value.retries.every(isAgentRetry) && value.children.every(isAgentHop);
 }
 function isAgentTrace(value) {
   if (!isObjectRecord2(value) || !Array.isArray(value.agents) || !Array.isArray(value.hops)) {
@@ -201909,7 +202112,7 @@ function loadAgentRunRecording(recordingFilePath, options = {}) {
 }
 
 // src/knowledge/knowledgeRuntime.ts
-var import_crypto10 = require("crypto");
+var import_crypto11 = require("crypto");
 var path35 = __toESM(require("path"));
 
 // src/knowledge/knowledgeIndexBuilder.ts
@@ -204507,7 +204710,7 @@ var extensionEnsurer;
 var KnowledgeRunCoordinator = class {
   constructor(options) {
     this.options = options;
-    this.runId = options.runId ?? `run-${(0, import_crypto10.randomUUID)()}`;
+    this.runId = options.runId ?? `run-${(0, import_crypto11.randomUUID)()}`;
     this.projectRoot = path35.resolve(findProjectRoot(options.startPath));
     this.snapshotId = options.snapshotId;
     this.ensureCurrent = options.ensureCurrent ?? extensionEnsurer ?? (async (projectRoot) => ensureKnowledgeIndexCurrent(projectRoot, { projectRoot }).snapshotId);
@@ -208899,7 +209102,7 @@ function redactUrl(url3, options) {
 }
 
 // src/cli/jsonRedaction.ts
-var import_crypto11 = require("crypto");
+var import_crypto12 = require("crypto");
 function redactAssertionForJson(assertion, redaction) {
   return {
     ...assertion,
@@ -208966,7 +209169,7 @@ function redactAgentRequestForJson(request, redaction) {
         }))
       }))
     } : void 0,
-    promptHash: (0, import_crypto11.createHash)("sha256").update(systemPrompt, "utf8").digest("hex")
+    promptHash: (0, import_crypto12.createHash)("sha256").update(systemPrompt, "utf8").digest("hex")
   };
 }
 function redactAgentHopForJson(hop, redaction) {
@@ -212576,6 +212779,7 @@ var ReplayCoordinator = class {
       ...hop.request.reasoningEffort ? { reasoningEffort: hop.request.reasoningEffort } : {},
       systemPrompt: hop.request.systemPrompt,
       maxTokens: hop.request.maxTokens,
+      ...hop.request.contextTokens !== void 0 ? { contextTokens: hop.request.contextTokens } : {},
       ...hop.request.maxInputTokens !== void 0 ? { maxInputTokens: hop.request.maxInputTokens } : {},
       tools: hop.request.tools.map((tool) => ({
         ...tool,
@@ -212869,7 +213073,11 @@ function addSourceDriftDiagnostics(recording, scopes, runtimeVariables, recordin
       );
     }
     const current = substituteVariables(comparable.systemPrompt, runtimeVariables);
-    if (/\{\{[^}]+\}\}/.test(current) || current === hop.request.systemPrompt) {
+    const recoverySuffix = `
+
+${TOOL_RESULT_RECOVERY_INSTRUCTIONS}`;
+    const recordedPrompt = hop.request.systemPrompt.endsWith(recoverySuffix) ? hop.request.systemPrompt.slice(0, -recoverySuffix.length) : hop.request.systemPrompt;
+    if (/\{\{[^}]+\}\}/.test(current) || current === recordedPrompt) {
       continue;
     }
     diagnostics.push(

package/dist/knowledgeIndexWorker.js

@@ -7647,6 +7647,7 @@ function splitHeaderDirective(value) {
 var LIMIT_DIRECTIVES = {
   max_tokens: "maxTokens",
   max_input_tokens: "maxInputTokens",
+  context_tokens: "contextTokens",
   max_depth: "maxDepth",
   max_invocations: "maxInvocations",
   max_turns: "maxTurns",
@@ -8190,6 +8191,7 @@ end mcp
         returns: pending.returns,
         maxTokens: pending.maxTokens?.value,
         maxInputTokens: pending.maxInputTokens?.value,
+        contextTokens: pending.contextTokens?.value,
         maxDepth: pending.maxDepth?.value,
         maxInvocations: pending.maxInvocations?.value,
         maxTurns: pending.maxTurns?.value,
@@ -9626,6 +9628,7 @@ var AGENT_LIMIT_CONFIG_KEYS = /* @__PURE__ */ new Set([
   "_comment",
   "max_tokens",
   "max_input_tokens",
+  "context_tokens",
   "max_depth",
   "max_invocations",
   "max_turns",
@@ -9635,7 +9638,7 @@ function hasOnlyKeys(value, allowedKeys) {
   return Object.keys(value).every((key) => allowedKeys.has(key));
 }
 function hasValidNornAgentLimits(value) {
-  return (value.max_tokens === void 0 || isPositiveInteger(value.max_tokens)) && (value.max_input_tokens === void 0 || isPositiveInteger(value.max_input_tokens)) && (value.max_depth === void 0 || isPositiveInteger(value.max_depth)) && (value.max_invocations === void 0 || isPositiveInteger(value.max_invocations)) && (value.max_turns === void 0 || isPositiveInteger(value.max_turns)) && (value.contract_retries === void 0 || isNonNegativeInteger(value.contract_retries));
+  return (value.max_tokens === void 0 || isPositiveInteger(value.max_tokens)) && (value.max_input_tokens === void 0 || isPositiveInteger(value.max_input_tokens)) && (value.context_tokens === void 0 || isPositiveInteger(value.context_tokens)) && (value.max_depth === void 0 || isPositiveInteger(value.max_depth)) && (value.max_invocations === void 0 || isPositiveInteger(value.max_invocations)) && (value.max_turns === void 0 || isPositiveInteger(value.max_turns)) && (value.contract_retries === void 0 || isNonNegativeInteger(value.contract_retries));
 }
 function isNornAgentProviderConfig(value) {
   if (!isObjectRecord(value)) {

package/package.json

@@ -2,7 +2,7 @@
   "name": "norn-cli",
   "displayName": "Norn",
   "description": "Build version-controlled agentic flows in VS Code and run them from the CLI or CI.",
-  "version": "4.5.1",
+  "version": "4.6.0",
   "publisher": "Norn-PeterKrustanov",
   "author": {
     "name": "Peter Krastanov"

package/schemas/norn.config.schema.json

@@ -84,6 +84,7 @@
         "version": 1,
         "agents": {
           "max_tokens": 16000,
+          "context_tokens": 64000,
           "max_depth": 5,
           "max_invocations": 25,
           "max_turns": 100,
@@ -93,7 +94,8 @@
           },
           "providers": {
             "local": {
-              "max_tokens": 4096
+              "max_tokens": 4096,
+              "context_tokens": 16000
             }
           }
         }
@@ -215,6 +217,7 @@
           "description": "Insert global agent limits, recording settings, and an optional provider-specific override.",
           "body": {
             "max_tokens": 16000,
+            "context_tokens": 64000,
             "max_depth": 5,
             "max_invocations": 25,
             "max_turns": 100,
@@ -224,7 +227,8 @@
             },
             "providers": {
               "${1:local}": {
-                "max_tokens": 4096
+                "max_tokens": 4096,
+                "context_tokens": 16000
               }
             }
           }
@@ -431,6 +435,11 @@
       "additionalProperties": false,
       "description": "Agent limits for one model provider. These override global agents settings and are overridden by directives on an individual agent.",
       "properties": {
+        "context_tokens": {
+          "type": "integer",
+          "minimum": 1,
+          "description": "Estimated input-context compaction budget per request. Excludes output tokens; choose below the model window allowing output headroom. This is a soft compaction budget, not a provider context-window guarantee."
+        },
         "_comment": {
           "$ref": "#/definitions/comment",
           "description": "Optional human-readable guidance ignored by Norn at runtime."
@@ -488,6 +497,11 @@
       "additionalProperties": false,
       "description": "Global recording behavior and global/provider-specific limits for agent runs. Per-agent .nornai or .nornagent directives override limit values.",
       "properties": {
+        "context_tokens": {
+          "type": "integer",
+          "minimum": 1,
+          "description": "Estimated input-context compaction budget per request. Defaults to 64000 (local: 16000). Ordinary context compacts from 80% toward 60%; recovery excerpts have a separate 10% allowance. Excludes output; set below the model window to leave output headroom. Prompts and the newest result are preserved even if they exceed this soft budget."
+        },
         "_comment": {
           "$ref": "#/definitions/comment",
           "description": "Optional human-readable guidance ignored by Norn at runtime."