@raquezha/noheadroom 0.0.2 → 0.2.0

package/dist/bridge.d.ts

@@ -1,3 +1,11 @@
 import type { AgentMessage, ApplyCompressionOptions, ApplyCompressionResult, CompressionMapping, CompressionPayload, OpenAIMessage } from "./types.js";
+type AnyMessage = AgentMessage & {
+    role?: string;
+    content?: unknown;
+    timestamp?: number;
+};
 export declare function buildCompressionPayload(messages: AgentMessage[], minMessageChars: number): CompressionPayload;
 export declare function applyCompressionResult(originalMessages: AgentMessage[], mappings: CompressionMapping[], compressedMessages: OpenAIMessage[], _options: ApplyCompressionOptions): ApplyCompressionResult;
+export declare function convertMessage(message: AnyMessage): OpenAIMessage | undefined;
+export declare function extractOpenAIText(message: OpenAIMessage): string;
+export {};

package/dist/bridge.js

@@ -8,8 +8,11 @@ export function buildCompressionPayload(messages, minMessageChars) {
         if (!converted)
             continue;
         const originalText = extractOpenAIText(converted);
-        const applyTo = source.role === "toolResult" && originalText.length >= minMessageChars ? "toolResult" : null;
-        if (applyTo)
+        // Allow any toolResult to be a candidate for compression if Headroom decides to shrink it.
+        // The minMessageChars threshold is primarily to avoid overhead for tiny messages,
+        // but we shouldn't block Headroom if it finds savings in slightly smaller ones.
+        const applyTo = source.role === "toolResult" ? "toolResult" : null;
+        if (applyTo && originalText.length >= minMessageChars)
             candidateCount++;
         mappings.push({ sourceIndex, message: converted, applyTo, originalText });
     }
@@ -51,7 +54,7 @@ export function applyCompressionResult(originalMessages, mappings, compressedMes
     }
     return { ok: true, messages: nextMessages, appliedMessages };
 }
-function convertMessage(message) {
+export function convertMessage(message) {
     if (!message.role || !STANDARD_COMPRESSIBLE_ROLES.has(message.role))
         return undefined;
     switch (message.role) {
@@ -142,7 +145,7 @@ function validateAlignedMessage(original, compressed) {
     }
     return { ok: true };
 }
-function extractOpenAIText(message) {
+export function extractOpenAIText(message) {
     if (message.role === "assistant")
         return typeof message.content === "string" ? message.content : "";
     if (message.role === "tool" || message.role === "system")

package/dist/index.d.ts

@@ -1,5 +1,6 @@
-import type { ExtensionAPI, ExtensionContext } from "@earendil-works/pi-coding-agent";
-import type { HeadroomConfig, HeadroomStats } from "./types.js";
+import type { ContextEvent, ExtensionAPI, ExtensionContext } from "@earendil-works/pi-coding-agent";
+import { HeadroomHttpClient } from "./client.js";
+import type { AgentMessage, HeadroomConfig, HeadroomStats } from "./types.js";
 interface HeadroomRuntimeState {
     enabled: boolean;
     proxyOnline: boolean | null;
@@ -7,13 +8,32 @@ interface HeadroomRuntimeState {
     proxyStartAttempted: boolean;
     remoteWarningShown: boolean;
     offlineWarningShown: boolean;
+    processing: boolean;
+    lastInputFingerprint: string | null;
+    lastOutputFingerprint: string | null;
+    lastCompressionTime: number;
     stats: HeadroomStats;
 }
+interface HeadroomRuntime {
+    pi: ExtensionAPI;
+    config: HeadroomConfig;
+    client: HeadroomHttpClient;
+    state: HeadroomRuntimeState;
+    refreshStatus(ctx: ExtensionContext): void;
+    updateHealth(ctx: ExtensionContext): Promise<boolean>;
+    ensureProxy(ctx: ExtensionContext): Promise<boolean>;
+}
 export default function headroomExtension(pi: ExtensionAPI): void;
+declare function handleContextCompression(runtime: HeadroomRuntime, event: ContextEvent, ctx: ExtensionContext): Promise<{
+    messages?: AgentMessage[];
+} | undefined>;
+declare function generateFingerprint(messages: AgentMessage[]): string;
 declare function isAbortOrTimeoutError(error: unknown): boolean;
 declare function renderFooterStatus(ctx: ExtensionContext, config: HeadroomConfig, state: HeadroomRuntimeState): string;
 export declare const __test__: {
     isAbortOrTimeoutError: typeof isAbortOrTimeoutError;
     renderFooterStatus: typeof renderFooterStatus;
+    generateFingerprint: typeof generateFingerprint;
+    handleContextCompression: typeof handleContextCompression;
 };
 export {};

package/dist/index.js

@@ -1,5 +1,5 @@
 import { Box, Text } from "@earendil-works/pi-tui";
-import { applyCompressionResult, buildCompressionPayload } from "./bridge.js";
+import { applyCompressionResult, buildCompressionPayload, convertMessage, extractOpenAIText, } from "./bridge.js";
 import { HeadroomHttpClient } from "./client.js";
 import { isRemoteBlocked, loadHeadroomConfig } from "./config.js";
 import { startPersistentHeadroomProxy } from "./proxy-manager.js";
@@ -49,6 +49,10 @@ function createRuntime(pi) {
         proxyStartAttempted: false,
         remoteWarningShown: false,
         offlineWarningShown: false,
+        processing: false,
+        lastInputFingerprint: null,
+        lastOutputFingerprint: null,
+        lastCompressionTime: 0,
         stats: { attempts: 0, applied: 0, guardSkips: 0, tokensSaved: 0 },
     };
     const runtime = {
@@ -154,7 +158,7 @@ function sleep(ms) {
 function registerCompressionRenderer(pi) {
     pi.registerMessageRenderer(COMPRESSION_ENTRY_TYPE, (message, { expanded }, theme) => {
         const details = message.details;
-        const content = typeof message.content === "string" ? message.content : "noheadroom compression applied";
+        const content = details?.summary ?? (typeof message.content === "string" ? message.content : "noheadroom compression applied");
         const box = new Box(1, 1, (text) => theme.bg("customMessageBg", text));
         let text = `${theme.fg("success", "🗜 noheadroom")} ${content}`;
         if (expanded && details) {
@@ -169,6 +173,23 @@ function registerCompressionRenderer(pi) {
     });
 }
 async function handleContextCompression(runtime, event, ctx) {
+    if (runtime.state.processing)
+        return undefined;
+    // Throttle: max 1 compression attempt per 3 seconds to kill infinite loops
+    const now = Date.now();
+    if (now - runtime.state.lastCompressionTime < 3000) {
+        return undefined;
+    }
+    // Content-based guards to prevent infinite recursion
+    const inputFingerprint = generateFingerprint(event.messages);
+    // 1. If this input matches our previous output, we already compressed it. Stop.
+    if (runtime.state.lastOutputFingerprint === inputFingerprint) {
+        return undefined;
+    }
+    // 2. If this input matches our previous input, it didn't change. Stop.
+    if (runtime.state.lastInputFingerprint === inputFingerprint) {
+        return undefined;
+    }
     if (shouldSkipBeforePayload(runtime, ctx))
         return undefined;
     const payload = buildCompressionPayload(event.messages, runtime.config.minMessageChars);
@@ -178,11 +199,15 @@ async function handleContextCompression(runtime, event, ctx) {
         void ensureProxyInBackground(runtime, ctx);
         return undefined;
     }
+    runtime.state.processing = true;
+    runtime.state.lastCompressionTime = now;
     runtime.state.stats.attempts++;
     try {
         const result = await runtime.client.compress(payload.messages, ctx.model?.id, ctx.signal);
         runtime.state.proxyOnline = true;
         if (!result.compressed || result.tokensSaved <= 0) {
+            // Even if no tokens saved, record the input so we don't keep trying
+            runtime.state.lastInputFingerprint = inputFingerprint;
             runtime.refreshStatus(ctx);
             return undefined;
         }
@@ -195,6 +220,9 @@ async function handleContextCompression(runtime, event, ctx) {
             runtime.refreshStatus(ctx);
             return undefined;
         }
+        // Store fingerprints to break the feedback loop
+        runtime.state.lastInputFingerprint = inputFingerprint;
+        runtime.state.lastOutputFingerprint = generateFingerprint(applied.messages);
         recordAppliedCompression(runtime.state.stats, result, applied.appliedMessages);
         announceAppliedCompression(runtime.pi, ctx, result, applied.appliedMessages);
         runtime.refreshStatus(ctx);
@@ -204,6 +232,18 @@ async function handleContextCompression(runtime, event, ctx) {
         recordCompressionError(runtime, ctx, error);
         return undefined;
     }
+    finally {
+        runtime.state.processing = false;
+    }
+}
+function generateFingerprint(messages) {
+    return messages
+        .map((m) => {
+        const converted = convertMessage(m);
+        const len = converted ? extractOpenAIText(converted).length : 0;
+        return `${m.role}:${len}`;
+    })
+        .join(",");
 }
 function shouldSkipBeforePayload(runtime, ctx) {
     if (!runtime.state.enabled)
@@ -235,20 +275,8 @@ function announceAppliedCompression(pi, ctx, result, appliedMessages) {
     const summary = `compressed ${result.tokensBefore.toLocaleString()} → ${result.tokensAfter.toLocaleString()} tokens (-${pct}%, saved ${result.tokensSaved.toLocaleString()}, messages ${appliedMessages})`;
     const line = `noheadroom: ${summary}`;
     ctx.ui.notify(line, "info");
-    pi.sendMessage({
-        customType: COMPRESSION_ENTRY_TYPE,
-        content: summary,
-        display: true,
-        details: {
-            tokensBefore: result.tokensBefore,
-            tokensAfter: result.tokensAfter,
-            tokensSaved: result.tokensSaved,
-            compressionRatio: result.compressionRatio,
-            appliedMessages,
-            transformsApplied: result.transformsApplied,
-            timestamp: Date.now(),
-        },
-    });
+    // Note: we no longer use pi.appendEntry or pi.sendMessage here because
+    // modifying the session history inside the context event triggers an infinite loop.
     // Non-interactive `pi -p` does not always show footer/status UI, so print an explicit proof line.
     process.stderr.write(`🗜 ${line}\n`);
 }
@@ -428,4 +456,6 @@ function parseSubcommand(args) {
 export const __test__ = {
     isAbortOrTimeoutError,
     renderFooterStatus,
+    generateFingerprint,
+    handleContextCompression,
 };

package/extensions/bridge.ts

@@ -28,8 +28,11 @@ export function buildCompressionPayload(messages: AgentMessage[], minMessageChar
 		if (!converted) continue;
 
 		const originalText = extractOpenAIText(converted);
-		const applyTo = source.role === "toolResult" && originalText.length >= minMessageChars ? "toolResult" : null;
-		if (applyTo) candidateCount++;
+		// Allow any toolResult to be a candidate for compression if Headroom decides to shrink it.
+		// The minMessageChars threshold is primarily to avoid overhead for tiny messages,
+		// but we shouldn't block Headroom if it finds savings in slightly smaller ones.
+		const applyTo = source.role === "toolResult" ? "toolResult" : null;
+		if (applyTo && originalText.length >= minMessageChars) candidateCount++;
 		mappings.push({ sourceIndex, message: converted, applyTo, originalText });
 	}
 
@@ -84,7 +87,7 @@ export function applyCompressionResult(
 	return { ok: true, messages: nextMessages, appliedMessages };
 }
 
-function convertMessage(message: AnyMessage): OpenAIMessage | undefined {
+export function convertMessage(message: AnyMessage): OpenAIMessage | undefined {
 	if (!message.role || !STANDARD_COMPRESSIBLE_ROLES.has(message.role)) return undefined;
 	switch (message.role) {
 		case "user":
@@ -176,7 +179,7 @@ function validateAlignedMessage(
 	return { ok: true };
 }
 
-function extractOpenAIText(message: OpenAIMessage): string {
+export function extractOpenAIText(message: OpenAIMessage): string {
 	if (message.role === "assistant") return typeof message.content === "string" ? message.content : "";
 	if (message.role === "tool" || message.role === "system") return message.content;
 	if (typeof message.content === "string") return message.content;

package/extensions/index.ts

@@ -1,6 +1,11 @@
 import type { ContextEvent, ExtensionAPI, ExtensionContext } from "@earendil-works/pi-coding-agent";
 import { Box, Text } from "@earendil-works/pi-tui";
-import { applyCompressionResult, buildCompressionPayload } from "./bridge.js";
+import {
+	applyCompressionResult,
+	buildCompressionPayload,
+	convertMessage,
+	extractOpenAIText,
+} from "./bridge.js";
 import { HeadroomHttpClient } from "./client.js";
 import { isRemoteBlocked, loadHeadroomConfig } from "./config.js";
 import { startPersistentHeadroomProxy } from "./proxy-manager.js";
@@ -18,6 +23,10 @@ interface HeadroomRuntimeState {
 	proxyStartAttempted: boolean;
 	remoteWarningShown: boolean;
 	offlineWarningShown: boolean;
+	processing: boolean;
+	lastInputFingerprint: string | null;
+	lastOutputFingerprint: string | null;
+	lastCompressionTime: number;
 	stats: HeadroomStats;
 }
 
@@ -93,6 +102,10 @@ function createRuntime(pi: ExtensionAPI): HeadroomRuntime {
 		proxyStartAttempted: false,
 		remoteWarningShown: false,
 		offlineWarningShown: false,
+		processing: false,
+		lastInputFingerprint: null,
+		lastOutputFingerprint: null,
+		lastCompressionTime: 0,
 		stats: { attempts: 0, applied: 0, guardSkips: 0, tokensSaved: 0 },
 	};
 
@@ -200,8 +213,8 @@ function sleep(ms: number): Promise<void> {
 
 function registerCompressionRenderer(pi: ExtensionAPI): void {
 	pi.registerMessageRenderer(COMPRESSION_ENTRY_TYPE, (message, { expanded }, theme) => {
-		const details = message.details as CompressionEntryDetails | undefined;
-		const content = typeof message.content === "string" ? message.content : "noheadroom compression applied";
+		const details = message.details as (CompressionEntryDetails & { summary?: string }) | undefined;
+		const content = details?.summary ?? (typeof message.content === "string" ? message.content : "noheadroom compression applied");
 		const box = new Box(1, 1, (text) => theme.bg("customMessageBg", text));
 		let text = `${theme.fg("success", "🗜 noheadroom")} ${content}`;
 		if (expanded && details) {
@@ -221,6 +234,27 @@ async function handleContextCompression(
 	event: ContextEvent,
 	ctx: ExtensionContext,
 ): Promise<{ messages?: AgentMessage[] } | undefined> {
+	if (runtime.state.processing) return undefined;
+
+	// Throttle: max 1 compression attempt per 3 seconds to kill infinite loops
+	const now = Date.now();
+	if (now - runtime.state.lastCompressionTime < 3000) {
+		return undefined;
+	}
+
+	// Content-based guards to prevent infinite recursion
+	const inputFingerprint = generateFingerprint(event.messages);
+
+	// 1. If this input matches our previous output, we already compressed it. Stop.
+	if (runtime.state.lastOutputFingerprint === inputFingerprint) {
+		return undefined;
+	}
+
+	// 2. If this input matches our previous input, it didn't change. Stop.
+	if (runtime.state.lastInputFingerprint === inputFingerprint) {
+		return undefined;
+	}
+
 	if (shouldSkipBeforePayload(runtime, ctx)) return undefined;
 	const payload = buildCompressionPayload(event.messages, runtime.config.minMessageChars);
 	if (payload.candidateCount === 0) return undefined;
@@ -229,11 +263,15 @@ async function handleContextCompression(
 		return undefined;
 	}
 
+	runtime.state.processing = true;
+	runtime.state.lastCompressionTime = now;
 	runtime.state.stats.attempts++;
 	try {
 		const result = await runtime.client.compress(payload.messages, ctx.model?.id, ctx.signal);
 		runtime.state.proxyOnline = true;
 		if (!result.compressed || result.tokensSaved <= 0) {
+			// Even if no tokens saved, record the input so we don't keep trying
+			runtime.state.lastInputFingerprint = inputFingerprint;
 			runtime.refreshStatus(ctx);
 			return undefined;
 		}
@@ -248,6 +286,10 @@ async function handleContextCompression(
 			return undefined;
 		}
 
+		// Store fingerprints to break the feedback loop
+		runtime.state.lastInputFingerprint = inputFingerprint;
+		runtime.state.lastOutputFingerprint = generateFingerprint(applied.messages);
+
 		recordAppliedCompression(runtime.state.stats, result, applied.appliedMessages);
 		announceAppliedCompression(runtime.pi, ctx, result, applied.appliedMessages);
 		runtime.refreshStatus(ctx);
@@ -255,9 +297,21 @@ async function handleContextCompression(
 	} catch (error) {
 		recordCompressionError(runtime, ctx, error);
 		return undefined;
+	} finally {
+		runtime.state.processing = false;
 	}
 }
 
+function generateFingerprint(messages: AgentMessage[]): string {
+	return messages
+		.map((m) => {
+			const converted = convertMessage(m);
+			const len = converted ? extractOpenAIText(converted).length : 0;
+			return `${m.role}:${len}`;
+		})
+		.join(",");
+}
+
 function shouldSkipBeforePayload(runtime: HeadroomRuntime, ctx: ExtensionContext): boolean {
 	if (!runtime.state.enabled) return true;
 	if (isRemoteBlocked(runtime.config)) {
@@ -295,20 +349,8 @@ function announceAppliedCompression(
 	const summary = `compressed ${result.tokensBefore.toLocaleString()} → ${result.tokensAfter.toLocaleString()} tokens (-${pct}%, saved ${result.tokensSaved.toLocaleString()}, messages ${appliedMessages})`;
 	const line = `noheadroom: ${summary}`;
 	ctx.ui.notify(line, "info");
-	pi.sendMessage({
-		customType: COMPRESSION_ENTRY_TYPE,
-		content: summary,
-		display: true,
-		details: {
-			tokensBefore: result.tokensBefore,
-			tokensAfter: result.tokensAfter,
-			tokensSaved: result.tokensSaved,
-			compressionRatio: result.compressionRatio,
-			appliedMessages,
-			transformsApplied: result.transformsApplied,
-			timestamp: Date.now(),
-		} satisfies CompressionEntryDetails,
-	});
+	// Note: we no longer use pi.appendEntry or pi.sendMessage here because
+	// modifying the session history inside the context event triggers an infinite loop.
 	// Non-interactive `pi -p` does not always show footer/status UI, so print an explicit proof line.
 	process.stderr.write(`🗜 ${line}\n`);
 }
@@ -524,4 +566,6 @@ function parseSubcommand(args: string): Subcommand {
 export const __test__ = {
 	isAbortOrTimeoutError,
 	renderFooterStatus,
+	generateFingerprint,
+	handleContextCompression,
 };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@raquezha/noheadroom",
-  "version": "0.0.2",
+  "version": "0.2.0",
   "description": "Personal Headroom compression bridge for Pi with Docker-first defaults and Pi tool-name adaptation",
   "main": "dist/index.js",
   "type": "module",
@@ -34,6 +34,13 @@
     "access": "public"
   },
   "types": "dist/index.d.ts",
+  "files": [
+    "dist",
+    "extensions",
+    "README.md",
+    "LICENSE",
+    "UPSTREAM-README.md"
+  ],
   "pi": {
     "extensions": [
       "extensions"

package/CHANGELOG.md

@@ -1,7 +0,0 @@
-# @raquezha/noheadroom
-
-## 0.0.2
-
-### Patch Changes
-
-- 0d65b37: Initial release of the Headroom compression bridge for Pi.

package/tsconfig.json

@@ -1,15 +0,0 @@
-{
-  "compilerOptions": {
-    "target": "ES2022",
-    "module": "NodeNext",
-    "moduleResolution": "NodeNext",
-    "declaration": true,
-    "outDir": "./dist",
-    "rootDir": "extensions",
-    "strict": true,
-    "esModuleInterop": true,
-    "skipLibCheck": true,
-    "forceConsistentCasingInFileNames": true
-  },
-  "include": ["extensions/**/*.ts"]
-}