feat(agent): opt-in tool-result context pruning

src/agents/pi-embedded-runner.ts

@@ -1,5 +1,7 @@
 import fs from "node:fs/promises";
 import os from "node:os";
+import path from "node:path";
+import { fileURLToPath } from "node:url";
 
 import type {
   AgentMessage,
@@ -40,7 +42,11 @@ import {
   markAuthProfileUsed,
 } from "./auth-profiles.js";
 import type { BashElevatedDefaults } from "./bash-tools.js";
-import { DEFAULT_MODEL, DEFAULT_PROVIDER } from "./defaults.js";
+import {
+  DEFAULT_CONTEXT_TOKENS,
+  DEFAULT_MODEL,
+  DEFAULT_PROVIDER,
+} from "./defaults.js";
 import {
   ensureAuthProfileStore,
   getApiKeyForModel,
@@ -67,6 +73,9 @@ import {
   extractAssistantThinking,
   formatReasoningMarkdown,
 } from "./pi-embedded-utils.js";
+import { setContextPruningRuntime } from "./pi-extensions/context-pruning/runtime.js";
+import { computeEffectiveSettings } from "./pi-extensions/context-pruning/settings.js";
+import { makeToolPrunablePredicate } from "./pi-extensions/context-pruning/tools.js";
 import { toToolDefinitions } from "./pi-tool-definition-adapter.js";
 import { createClawdbotCodingTools } from "./pi-tools.js";
 import { resolveSandboxContext } from "./sandbox.js";
@@ -82,6 +91,84 @@ import { buildAgentSystemPromptAppend } from "./system-prompt.js";
 import { normalizeUsage, type UsageLike } from "./usage.js";
 import { loadWorkspaceBootstrapFiles } from "./workspace.js";
 
+// Optional features can be implemented as Pi extensions that run in the same Node process.
+// We configure context pruning per-session via a WeakMap registry keyed by the SessionManager instance.
+
+function resolvePiExtensionPath(id: string): string {
+  const self = fileURLToPath(import.meta.url);
+  const dir = path.dirname(self);
+  // In dev this file is `.ts` (tsx), in production it's `.js`.
+  const ext = path.extname(self) === ".ts" ? "ts" : "js";
+  return path.join(dir, "pi-extensions", `${id}.${ext}`);
+}
+
+function resolveContextWindowTokens(params: {
+  cfg: ClawdbotConfig | undefined;
+  provider: string;
+  modelId: string;
+  model: Model<Api> | undefined;
+}): number {
+  const fromModel =
+    typeof params.model?.contextWindow === "number" &&
+    Number.isFinite(params.model.contextWindow) &&
+    params.model.contextWindow > 0
+      ? params.model.contextWindow
+      : undefined;
+  if (fromModel) return fromModel;
+
+  const fromModelsConfig = (() => {
+    const providers = params.cfg?.models?.providers as
+      | Record<
+          string,
+          { models?: Array<{ id?: string; contextWindow?: number }> }
+        >
+      | undefined;
+    const providerEntry = providers?.[params.provider];
+    const models = Array.isArray(providerEntry?.models)
+      ? providerEntry.models
+      : [];
+    const match = models.find((m) => m?.id === params.modelId);
+    return typeof match?.contextWindow === "number" && match.contextWindow > 0
+      ? match.contextWindow
+      : undefined;
+  })();
+  if (fromModelsConfig) return fromModelsConfig;
+
+  const fromAgentConfig =
+    typeof params.cfg?.agent?.contextTokens === "number" &&
+    Number.isFinite(params.cfg.agent.contextTokens) &&
+    params.cfg.agent.contextTokens > 0
+      ? Math.floor(params.cfg.agent.contextTokens)
+      : undefined;
+  if (fromAgentConfig) return fromAgentConfig;
+
+  return DEFAULT_CONTEXT_TOKENS;
+}
+
+function buildContextPruningExtension(params: {
+  cfg: ClawdbotConfig | undefined;
+  sessionManager: SessionManager;
+  provider: string;
+  modelId: string;
+  model: Model<Api> | undefined;
+}): { additionalExtensionPaths?: string[] } {
+  const raw = params.cfg?.agent?.contextPruning;
+  if (raw?.mode !== "adaptive" && raw?.mode !== "aggressive") return {};
+
+  const settings = computeEffectiveSettings(raw);
+  if (!settings) return {};
+
+  setContextPruningRuntime(params.sessionManager, {
+    settings,
+    contextWindowTokens: resolveContextWindowTokens(params),
+    isToolPrunable: makeToolPrunablePredicate(settings.tools),
+  });
+
+  return {
+    additionalExtensionPaths: [resolvePiExtensionPath("context-pruning")],
+  };
+}
+
 export type EmbeddedPiAgentMeta = {
   sessionId: string;
   provider: string;
@@ -578,13 +665,22 @@ export async function compactEmbeddedPiSession(params: {
           effectiveWorkspace,
           agentDir,
         );
+        const pruning = buildContextPruningExtension({
+          cfg: params.config,
+          sessionManager,
+          provider,
+          modelId,
+          model,
+        });
+        const additionalExtensionPaths = pruning.additionalExtensionPaths;
 
         const { builtInTools, customTools } = splitSdkTools({
           tools,
           sandboxEnabled: !!sandbox?.enabled,
         });
 
-        const { session } = await createAgentSession({
+        let session: Awaited<ReturnType<typeof createAgentSession>>["session"];
+        ({ session } = await createAgentSession({
           cwd: resolvedWorkspace,
           agentDir,
           authStorage,
@@ -598,7 +694,8 @@ export async function compactEmbeddedPiSession(params: {
           settingsManager,
           skills: promptSkills,
           contextFiles,
-        });
+          additionalExtensionPaths,
+        }));
 
         try {
           const prior = await sanitizeSessionMessagesImages(
@@ -887,13 +984,24 @@ export async function runEmbeddedPiAgent(params: {
             effectiveWorkspace,
             agentDir,
           );
+          const pruning = buildContextPruningExtension({
+            cfg: params.config,
+            sessionManager,
+            provider,
+            modelId,
+            model,
+          });
+          const additionalExtensionPaths = pruning.additionalExtensionPaths;
 
           const { builtInTools, customTools } = splitSdkTools({
             tools,
             sandboxEnabled: !!sandbox?.enabled,
           });
 
-          const { session } = await createAgentSession({
+          let session: Awaited<
+            ReturnType<typeof createAgentSession>
+          >["session"];
+          ({ session } = await createAgentSession({
             cwd: resolvedWorkspace,
             agentDir,
             authStorage,
@@ -909,8 +1017,10 @@ export async function runEmbeddedPiAgent(params: {
             settingsManager,
             skills: promptSkills,
             contextFiles,
-          });
+            additionalExtensionPaths,
+          }));
 
+          try {
             const prior = await sanitizeSessionMessagesImages(
               session.messages,
               "session:history",
@@ -918,6 +1028,10 @@ export async function runEmbeddedPiAgent(params: {
             if (prior.length > 0) {
               session.agent.replaceMessages(prior);
             }
+          } catch (err) {
+            session.dispose();
+            throw err;
+          }
           let aborted = Boolean(params.abortSignal?.aborted);
           let timedOut = false;
           const abortRun = (isTimeout = false) => {
@@ -925,7 +1039,9 @@ export async function runEmbeddedPiAgent(params: {
             if (isTimeout) timedOut = true;
             void session.abort();
           };
-          const subscription = subscribeEmbeddedPiSession({
+          let subscription: ReturnType<typeof subscribeEmbeddedPiSession>;
+          try {
+            subscription = subscribeEmbeddedPiSession({
               session,
               runId: params.runId,
               verboseLevel: params.verboseLevel,
@@ -940,6 +1056,10 @@ export async function runEmbeddedPiAgent(params: {
               onAgentEvent: params.onAgentEvent,
               enforceFinalTag: params.enforceFinalTag,
             });
+          } catch (err) {
+            session.dispose();
+            throw err;
+          }
           const {
             assistantTexts,
             toolMetas,

src/agents/pi-extensions/context-pruning.ts

@@ -0,0 +1,19 @@
+/**
+ * Opt-in context pruning (“microcompact”-style) for Pi sessions.
+ *
+ * This only affects the in-memory context for the current request; it does not rewrite session
+ * history persisted on disk.
+ */
+
+export { default } from "./context-pruning/extension.js";
+
+export { pruneContextMessages } from "./context-pruning/pruner.js";
+export type {
+  ContextPruningConfig,
+  ContextPruningToolMatch,
+  EffectiveContextPruningSettings,
+} from "./context-pruning/settings.js";
+export {
+  computeEffectiveSettings,
+  DEFAULT_CONTEXT_PRUNING_SETTINGS,
+} from "./context-pruning/settings.js";

src/agents/pi-extensions/context-pruning/extension.ts

@@ -0,0 +1,27 @@
+import type { AgentMessage } from "@mariozechner/pi-agent-core";
+import type {
+  ContextEvent,
+  ExtensionAPI,
+  ExtensionContext,
+} from "@mariozechner/pi-coding-agent";
+
+import { pruneContextMessages } from "./pruner.js";
+import { getContextPruningRuntime } from "./runtime.js";
+
+export default function contextPruningExtension(api: ExtensionAPI): void {
+  api.on("context", (event: ContextEvent, ctx: ExtensionContext) => {
+    const runtime = getContextPruningRuntime(ctx.sessionManager);
+    if (!runtime) return undefined;
+
+    const next = pruneContextMessages({
+      messages: event.messages as AgentMessage[],
+      settings: runtime.settings,
+      ctx,
+      isToolPrunable: runtime.isToolPrunable,
+      contextWindowTokensOverride: runtime.contextWindowTokens ?? undefined,
+    });
+
+    if (next === event.messages) return undefined;
+    return { messages: next };
+  });
+}

src/agents/pi-extensions/context-pruning/pruner.ts

@@ -0,0 +1,310 @@
+import type { AgentMessage } from "@mariozechner/pi-agent-core";
+import type {
+  ImageContent,
+  TextContent,
+  ToolResultMessage,
+} from "@mariozechner/pi-ai";
+import type { ExtensionContext } from "@mariozechner/pi-coding-agent";
+
+import type { EffectiveContextPruningSettings } from "./settings.js";
+import { makeToolPrunablePredicate } from "./tools.js";
+
+const CHARS_PER_TOKEN_ESTIMATE = 4;
+// We currently skip pruning tool results that contain images. Still, we count them (approx.) so
+// we start trimming prunable tool results earlier when image-heavy context is consuming the window.
+const IMAGE_CHAR_ESTIMATE = 8_000;
+
+function asText(text: string): TextContent {
+  return { type: "text", text };
+}
+
+function collectTextSegments(
+  content: ReadonlyArray<TextContent | ImageContent>,
+): string[] {
+  const parts: string[] = [];
+  for (const block of content) {
+    if (block.type === "text") parts.push(block.text);
+  }
+  return parts;
+}
+
+function estimateJoinedTextLength(parts: string[]): number {
+  if (parts.length === 0) return 0;
+  let len = 0;
+  for (const p of parts) len += p.length;
+  // Joined with "\n" separators between blocks.
+  len += Math.max(0, parts.length - 1);
+  return len;
+}
+
+function takeHeadFromJoinedText(parts: string[], maxChars: number): string {
+  if (maxChars <= 0 || parts.length === 0) return "";
+  let remaining = maxChars;
+  let out = "";
+  for (let i = 0; i < parts.length && remaining > 0; i++) {
+    if (i > 0) {
+      out += "\n";
+      remaining -= 1;
+      if (remaining <= 0) break;
+    }
+    const p = parts[i];
+    if (p.length <= remaining) {
+      out += p;
+      remaining -= p.length;
+    } else {
+      out += p.slice(0, remaining);
+      remaining = 0;
+    }
+  }
+  return out;
+}
+
+function takeTailFromJoinedText(parts: string[], maxChars: number): string {
+  if (maxChars <= 0 || parts.length === 0) return "";
+  let remaining = maxChars;
+  const out: string[] = [];
+  for (let i = parts.length - 1; i >= 0 && remaining > 0; i--) {
+    const p = parts[i];
+    if (p.length <= remaining) {
+      out.push(p);
+      remaining -= p.length;
+    } else {
+      out.push(p.slice(p.length - remaining));
+      remaining = 0;
+      break;
+    }
+    if (remaining > 0 && i > 0) {
+      out.push("\n");
+      remaining -= 1;
+    }
+  }
+  out.reverse();
+  return out.join("");
+}
+
+function hasImageBlocks(
+  content: ReadonlyArray<TextContent | ImageContent>,
+): boolean {
+  for (const block of content) {
+    if (block.type === "image") return true;
+  }
+  return false;
+}
+
+function estimateMessageChars(message: AgentMessage): number {
+  if (message.role === "user") {
+    const content = message.content;
+    if (typeof content === "string") return content.length;
+    let chars = 0;
+    for (const b of content) {
+      if (b.type === "text") chars += b.text.length;
+      if (b.type === "image") chars += IMAGE_CHAR_ESTIMATE;
+    }
+    return chars;
+  }
+
+  if (message.role === "assistant") {
+    let chars = 0;
+    for (const b of message.content) {
+      if (b.type === "text") chars += b.text.length;
+      if (b.type === "thinking") chars += b.thinking.length;
+      if (b.type === "toolCall") {
+        try {
+          chars += JSON.stringify(b.arguments ?? {}).length;
+        } catch {
+          chars += 128;
+        }
+      }
+    }
+    return chars;
+  }
+
+  if (message.role === "toolResult") {
+    let chars = 0;
+    for (const b of message.content) {
+      if (b.type === "text") chars += b.text.length;
+      if (b.type === "image") chars += IMAGE_CHAR_ESTIMATE;
+    }
+    return chars;
+  }
+
+  return 256;
+}
+
+function estimateContextChars(messages: AgentMessage[]): number {
+  return messages.reduce((sum, m) => sum + estimateMessageChars(m), 0);
+}
+
+function findAssistantCutoffIndex(
+  messages: AgentMessage[],
+  keepLastAssistants: number,
+): number | null {
+  // keepLastAssistants <= 0 => everything is potentially prunable.
+  if (keepLastAssistants <= 0) return messages.length;
+
+  let remaining = keepLastAssistants;
+  for (let i = messages.length - 1; i >= 0; i--) {
+    if (messages[i]?.role !== "assistant") continue;
+    remaining--;
+    if (remaining === 0) return i;
+  }
+
+  // Not enough assistant messages to establish a protected tail.
+  return null;
+}
+
+function softTrimToolResultMessage(params: {
+  msg: ToolResultMessage;
+  settings: EffectiveContextPruningSettings;
+}): ToolResultMessage | null {
+  const { msg, settings } = params;
+  // Ignore image tool results for now: these are often directly relevant and hard to partially prune safely.
+  if (hasImageBlocks(msg.content)) return null;
+
+  const parts = collectTextSegments(msg.content);
+  const rawLen = estimateJoinedTextLength(parts);
+  if (rawLen <= settings.softTrim.maxChars) return null;
+
+  const headChars = Math.max(0, settings.softTrim.headChars);
+  const tailChars = Math.max(0, settings.softTrim.tailChars);
+  if (headChars + tailChars >= rawLen) return null;
+
+  const head = takeHeadFromJoinedText(parts, headChars);
+  const tail = takeTailFromJoinedText(parts, tailChars);
+  const trimmed = `${head}
+...
+${tail}`;
+
+  const note = `
+
+[Tool result trimmed: kept first ${headChars} chars and last ${tailChars} chars of ${rawLen} chars.]`;
+
+  return { ...msg, content: [asText(trimmed + note)] };
+}
+
+export function pruneContextMessages(params: {
+  messages: AgentMessage[];
+  settings: EffectiveContextPruningSettings;
+  ctx: Pick<ExtensionContext, "model">;
+  isToolPrunable?: (toolName: string) => boolean;
+  contextWindowTokensOverride?: number;
+}): AgentMessage[] {
+  const { messages, settings, ctx } = params;
+  const contextWindowTokens =
+    typeof params.contextWindowTokensOverride === "number" &&
+    Number.isFinite(params.contextWindowTokensOverride) &&
+    params.contextWindowTokensOverride > 0
+      ? params.contextWindowTokensOverride
+      : ctx.model?.contextWindow;
+  if (!contextWindowTokens || contextWindowTokens <= 0) return messages;
+
+  const charWindow = contextWindowTokens * CHARS_PER_TOKEN_ESTIMATE;
+  if (charWindow <= 0) return messages;
+
+  const cutoffIndex = findAssistantCutoffIndex(
+    messages,
+    settings.keepLastAssistants,
+  );
+  if (cutoffIndex === null) return messages;
+
+  const isToolPrunable =
+    params.isToolPrunable ?? makeToolPrunablePredicate(settings.tools);
+
+  if (settings.mode === "aggressive") {
+    let next: AgentMessage[] | null = null;
+
+    for (let i = 0; i < cutoffIndex; i++) {
+      const msg = messages[i];
+      if (!msg || msg.role !== "toolResult") continue;
+      if (!isToolPrunable(msg.toolName)) continue;
+      if (hasImageBlocks(msg.content)) {
+        continue;
+      }
+
+      const alreadyCleared =
+        msg.content.length === 1 &&
+        msg.content[0]?.type === "text" &&
+        msg.content[0].text === settings.hardClear.placeholder;
+      if (alreadyCleared) continue;
+
+      const cleared: ToolResultMessage = {
+        ...msg,
+        content: [asText(settings.hardClear.placeholder)],
+      };
+      if (!next) next = messages.slice();
+      next[i] = cleared as unknown as AgentMessage;
+    }
+
+    return next ?? messages;
+  }
+
+  const totalCharsBefore = estimateContextChars(messages);
+  let totalChars = totalCharsBefore;
+  let ratio = totalChars / charWindow;
+  if (ratio < settings.softTrimRatio) {
+    return messages;
+  }
+
+  const prunableToolIndexes: number[] = [];
+  let next: AgentMessage[] | null = null;
+
+  for (let i = 0; i < cutoffIndex; i++) {
+    const msg = messages[i];
+    if (!msg || msg.role !== "toolResult") continue;
+    if (!isToolPrunable(msg.toolName)) continue;
+    if (hasImageBlocks(msg.content)) {
+      continue;
+    }
+    prunableToolIndexes.push(i);
+
+    const updated = softTrimToolResultMessage({
+      msg: msg as unknown as ToolResultMessage,
+      settings,
+    });
+    if (!updated) continue;
+
+    const beforeChars = estimateMessageChars(msg);
+    const afterChars = estimateMessageChars(updated as unknown as AgentMessage);
+    totalChars += afterChars - beforeChars;
+    if (!next) next = messages.slice();
+    next[i] = updated as unknown as AgentMessage;
+  }
+
+  const outputAfterSoftTrim = next ?? messages;
+  ratio = totalChars / charWindow;
+  if (ratio < settings.hardClearRatio) {
+    return outputAfterSoftTrim;
+  }
+  if (!settings.hardClear.enabled) {
+    return outputAfterSoftTrim;
+  }
+
+  let prunableToolChars = 0;
+  for (const i of prunableToolIndexes) {
+    const msg = outputAfterSoftTrim[i];
+    if (!msg || msg.role !== "toolResult") continue;
+    prunableToolChars += estimateMessageChars(msg);
+  }
+  if (prunableToolChars < settings.minPrunableToolChars) {
+    return outputAfterSoftTrim;
+  }
+
+  for (const i of prunableToolIndexes) {
+    if (ratio < settings.hardClearRatio) break;
+    const msg = (next ?? messages)[i];
+    if (!msg || msg.role !== "toolResult") continue;
+
+    const beforeChars = estimateMessageChars(msg);
+    const cleared: ToolResultMessage = {
+      ...msg,
+      content: [asText(settings.hardClear.placeholder)],
+    };
+    if (!next) next = messages.slice();
+    next[i] = cleared as unknown as AgentMessage;
+    const afterChars = estimateMessageChars(cleared as unknown as AgentMessage);
+    totalChars += afterChars - beforeChars;
+    ratio = totalChars / charWindow;
+  }
+
+  return next ?? messages;
+}

src/agents/pi-extensions/context-pruning/runtime.ts

@@ -0,0 +1,39 @@
+import type { EffectiveContextPruningSettings } from "./settings.js";
+
+export type ContextPruningRuntimeValue = {
+  settings: EffectiveContextPruningSettings;
+  contextWindowTokens?: number | null;
+  isToolPrunable: (toolName: string) => boolean;
+};
+
+// Session-scoped runtime registry keyed by object identity.
+// Important: this relies on Pi passing the same SessionManager object instance into
+// ExtensionContext (ctx.sessionManager) that we used when calling setContextPruningRuntime.
+const REGISTRY = new WeakMap<object, ContextPruningRuntimeValue>();
+
+export function setContextPruningRuntime(
+  sessionManager: unknown,
+  value: ContextPruningRuntimeValue | null,
+): void {
+  if (!sessionManager || typeof sessionManager !== "object") {
+    return;
+  }
+
+  const key = sessionManager as object;
+  if (value === null) {
+    REGISTRY.delete(key);
+    return;
+  }
+
+  REGISTRY.set(key, value);
+}
+
+export function getContextPruningRuntime(
+  sessionManager: unknown,
+): ContextPruningRuntimeValue | null {
+  if (!sessionManager || typeof sessionManager !== "object") {
+    return null;
+  }
+
+  return REGISTRY.get(sessionManager as object) ?? null;
+}

src/agents/pi-extensions/context-pruning/settings.ts

@@ -0,0 +1,135 @@
+export type ContextPruningToolMatch = {
+  allow?: string[];
+  deny?: string[];
+};
+
+export type ContextPruningMode = "off" | "adaptive" | "aggressive";
+
+export type ContextPruningConfig = {
+  mode?: ContextPruningMode;
+  keepLastAssistants?: number;
+  softTrimRatio?: number;
+  hardClearRatio?: number;
+  minPrunableToolChars?: number;
+  tools?: ContextPruningToolMatch;
+  softTrim?: {
+    maxChars?: number;
+    headChars?: number;
+    tailChars?: number;
+  };
+  hardClear?: {
+    enabled?: boolean;
+    placeholder?: string;
+  };
+};
+
+export type EffectiveContextPruningSettings = {
+  mode: Exclude<ContextPruningMode, "off">;
+  keepLastAssistants: number;
+  softTrimRatio: number;
+  hardClearRatio: number;
+  minPrunableToolChars: number;
+  tools: ContextPruningToolMatch;
+  softTrim: {
+    maxChars: number;
+    headChars: number;
+    tailChars: number;
+  };
+  hardClear: {
+    enabled: boolean;
+    placeholder: string;
+  };
+};
+
+export const DEFAULT_CONTEXT_PRUNING_SETTINGS: EffectiveContextPruningSettings =
+  {
+    mode: "adaptive",
+    keepLastAssistants: 3,
+    softTrimRatio: 0.3,
+    hardClearRatio: 0.5,
+    minPrunableToolChars: 50_000,
+    tools: {},
+    softTrim: {
+      maxChars: 4_000,
+      headChars: 1_500,
+      tailChars: 1_500,
+    },
+    hardClear: {
+      enabled: true,
+      placeholder: "[Old tool result content cleared]",
+    },
+  };
+
+export function computeEffectiveSettings(
+  raw: unknown,
+): EffectiveContextPruningSettings | null {
+  if (!raw || typeof raw !== "object") return null;
+  const cfg = raw as ContextPruningConfig;
+  if (cfg.mode !== "adaptive" && cfg.mode !== "aggressive") return null;
+
+  const s: EffectiveContextPruningSettings = structuredClone(
+    DEFAULT_CONTEXT_PRUNING_SETTINGS,
+  );
+  s.mode = cfg.mode;
+
+  if (
+    typeof cfg.keepLastAssistants === "number" &&
+    Number.isFinite(cfg.keepLastAssistants)
+  ) {
+    s.keepLastAssistants = Math.max(0, Math.floor(cfg.keepLastAssistants));
+  }
+  if (
+    typeof cfg.softTrimRatio === "number" &&
+    Number.isFinite(cfg.softTrimRatio)
+  ) {
+    s.softTrimRatio = Math.min(1, Math.max(0, cfg.softTrimRatio));
+  }
+  if (
+    typeof cfg.hardClearRatio === "number" &&
+    Number.isFinite(cfg.hardClearRatio)
+  ) {
+    s.hardClearRatio = Math.min(1, Math.max(0, cfg.hardClearRatio));
+  }
+  if (
+    typeof cfg.minPrunableToolChars === "number" &&
+    Number.isFinite(cfg.minPrunableToolChars)
+  ) {
+    s.minPrunableToolChars = Math.max(0, Math.floor(cfg.minPrunableToolChars));
+  }
+  if (cfg.tools) {
+    s.tools = cfg.tools;
+  }
+  if (cfg.softTrim) {
+    if (
+      typeof cfg.softTrim.maxChars === "number" &&
+      Number.isFinite(cfg.softTrim.maxChars)
+    ) {
+      s.softTrim.maxChars = Math.max(0, Math.floor(cfg.softTrim.maxChars));
+    }
+    if (
+      typeof cfg.softTrim.headChars === "number" &&
+      Number.isFinite(cfg.softTrim.headChars)
+    ) {
+      s.softTrim.headChars = Math.max(0, Math.floor(cfg.softTrim.headChars));
+    }
+    if (
+      typeof cfg.softTrim.tailChars === "number" &&
+      Number.isFinite(cfg.softTrim.tailChars)
+    ) {
+      s.softTrim.tailChars = Math.max(0, Math.floor(cfg.softTrim.tailChars));
+    }
+  }
+  if (cfg.hardClear) {
+    if (s.mode === "adaptive" && typeof cfg.hardClear.enabled === "boolean") {
+      s.hardClear.enabled = cfg.hardClear.enabled;
+    }
+    if (
+      typeof cfg.hardClear.placeholder === "string" &&
+      cfg.hardClear.placeholder.trim()
+    ) {
+      s.hardClear.placeholder = cfg.hardClear.placeholder.trim();
+    }
+  }
+
+  return s;
+}

src/agents/pi-extensions/context-pruning/tools.ts

@@ -0,0 +1,46 @@
+import type { ContextPruningToolMatch } from "./settings.js";
+
+function normalizePatterns(patterns?: string[]): string[] {
+  if (!Array.isArray(patterns)) return [];
+  return patterns.map((p) => String(p ?? "").trim()).filter(Boolean);
+}
+
+type CompiledPattern =
+  | { kind: "all" }
+  | { kind: "exact"; value: string }
+  | { kind: "regex"; value: RegExp };
+
+function compilePattern(pattern: string): CompiledPattern {
+  if (pattern === "*") return { kind: "all" };
+  if (!pattern.includes("*")) return { kind: "exact", value: pattern };
+
+  const escaped = pattern.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
+  const re = new RegExp(`^${escaped.replaceAll("\\*", ".*")}$`);
+  return { kind: "regex", value: re };
+}
+
+function compilePatterns(patterns?: string[]): CompiledPattern[] {
+  return normalizePatterns(patterns).map(compilePattern);
+}
+
+function matchesAny(toolName: string, patterns: CompiledPattern[]): boolean {
+  for (const p of patterns) {
+    if (p.kind === "all") return true;
+    if (p.kind === "exact" && toolName === p.value) return true;
+    if (p.kind === "regex" && p.value.test(toolName)) return true;
+  }
+  return false;
+}
+
+export function makeToolPrunablePredicate(
+  match: ContextPruningToolMatch,
+): (toolName: string) => boolean {
+  const deny = compilePatterns(match.deny);
+  const allow = compilePatterns(match.allow);
+
+  return (toolName: string) => {
+    if (matchesAny(toolName, deny)) return false;
+    if (allow.length === 0) return true;
+    return matchesAny(toolName, allow);
+  };
+}

src/config/types.ts

@@ -850,6 +850,27 @@ export type AgentModelListConfig = {
   fallbacks?: string[];
 };
 
+export type AgentContextPruningConfig = {
+  mode?: "off" | "adaptive" | "aggressive";
+  keepLastAssistants?: number;
+  softTrimRatio?: number;
+  hardClearRatio?: number;
+  minPrunableToolChars?: number;
+  tools?: {
+    allow?: string[];
+    deny?: string[];
+  };
+  softTrim?: {
+    maxChars?: number;
+    headChars?: number;
+    tailChars?: number;
+  };
+  hardClear?: {
+    enabled?: boolean;
+    placeholder?: string;
+  };
+};
+
 export type ClawdbotConfig = {
   auth?: AuthConfig;
   env?: {
@@ -895,6 +916,8 @@ export type ClawdbotConfig = {
     userTimezone?: string;
     /** Optional display-only context window override (used for % in status UIs). */
     contextTokens?: number;
+    /** Opt-in: prune old tool results from the LLM context to reduce token usage. */
+    contextPruning?: AgentContextPruningConfig;
     /** Default thinking level when no /think directive is present. */
     thinkingDefault?: "off" | "minimal" | "low" | "medium" | "high";
     /** Default verbose level when no /verbose directive is present. */

src/config/zod-schema.ts

@@ -513,6 +513,40 @@ export const ClawdbotSchema = z.object({
       skipBootstrap: z.boolean().optional(),
       userTimezone: z.string().optional(),
       contextTokens: z.number().int().positive().optional(),
+      contextPruning: z
+        .object({
+          mode: z
+            .union([
+              z.literal("off"),
+              z.literal("adaptive"),
+              z.literal("aggressive"),
+            ])
+            .optional(),
+          keepLastAssistants: z.number().int().nonnegative().optional(),
+          softTrimRatio: z.number().min(0).max(1).optional(),
+          hardClearRatio: z.number().min(0).max(1).optional(),
+          minPrunableToolChars: z.number().int().nonnegative().optional(),
+          tools: z
+            .object({
+              allow: z.array(z.string()).optional(),
+              deny: z.array(z.string()).optional(),
+            })
+            .optional(),
+          softTrim: z
+            .object({
+              maxChars: z.number().int().nonnegative().optional(),
+              headChars: z.number().int().nonnegative().optional(),
+              tailChars: z.number().int().nonnegative().optional(),
+            })
+            .optional(),
+          hardClear: z
+            .object({
+              enabled: z.boolean().optional(),
+              placeholder: z.string().optional(),
+            })
+            .optional(),
+        })
+        .optional(),
       tools: z
         .object({
           allow: z.array(z.string()).optional(),