context-compress 2026.3.22 → 2026.6.0

package/dist/util/fetch-code.d.ts

@@ -0,0 +1,10 @@
+/**
+ * Build a self-contained JS snippet that fetches a URL, converts HTML→markdown,
+ * and prints the result. Runs inside the sandbox subprocess.
+ *
+ * When `resolvedIp` is provided, the URL is rewritten to connect to that IP
+ * with the original Host header preserved — defeats DNS rebinding (TOCTOU)
+ * between the validation step and the actual fetch.
+ */
+export declare function buildFetchCode(url: string, resolvedIp?: string | null): string;
+//# sourceMappingURL=fetch-code.d.ts.map

package/dist/util/fetch-code.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"fetch-code.d.ts","sourceRoot":"","sources":["../../src/util/fetch-code.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AACH,wBAAgB,cAAc,CAAC,GAAG,EAAE,MAAM,EAAE,UAAU,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,MAAM,CA6E9E"}

package/dist/util/fetch-code.js

@@ -0,0 +1,87 @@
+/**
+ * Build a self-contained JS snippet that fetches a URL, converts HTML→markdown,
+ * and prints the result. Runs inside the sandbox subprocess.
+ *
+ * When `resolvedIp` is provided, the URL is rewritten to connect to that IP
+ * with the original Host header preserved — defeats DNS rebinding (TOCTOU)
+ * between the validation step and the actual fetch.
+ */
+export function buildFetchCode(url, resolvedIp) {
+    let fetchSetup;
+    if (resolvedIp) {
+        const pinnedUrl = new URL(url);
+        const originalHost = pinnedUrl.host;
+        // URL.hostname setter requires IPv6 literals to be bracketed; raw forms
+        // like "2001:db8::1" parse incorrectly (the first ":" is treated as a
+        // port delimiter). Detect IPv6 by colon-presence and wrap.
+        const hostnameValue = resolvedIp.includes(":") && !resolvedIp.startsWith("[") ? `[${resolvedIp}]` : resolvedIp;
+        pinnedUrl.hostname = hostnameValue;
+        fetchSetup = `
+const url = ${JSON.stringify(pinnedUrl.toString())};
+const resp = await fetch(url, { headers: { 'Host': ${JSON.stringify(originalHost)} }, redirect: 'error' });`;
+    }
+    else {
+        fetchSetup = `
+const url = ${JSON.stringify(url)};
+const resp = await fetch(url, { redirect: 'error' });`;
+    }
+    return `${fetchSetup}
+if (!resp.ok) { console.error("HTTP " + resp.status); process.exit(1); }
+const cl = resp.headers.get('content-length');
+if (cl && parseInt(cl, 10) > 10 * 1024 * 1024) {
+    console.error("Response too large: " + cl + " bytes"); process.exit(1);
+}
+const html = await resp.text();
+if (html.length > 10 * 1024 * 1024) {
+    console.error("Response body too large: " + html.length + " chars"); process.exit(1);
+}
+
+// Strip unwanted tags
+let md = html
+  .replace(/<script[^>]*>[\\s\\S]*?<\\/script>/gi, "")
+  .replace(/<style[^>]*>[\\s\\S]*?<\\/style>/gi, "")
+  .replace(/<nav[^>]*>[\\s\\S]*?<\\/nav>/gi, "")
+  .replace(/<header[^>]*>[\\s\\S]*?<\\/header>/gi, "")
+  .replace(/<footer[^>]*>[\\s\\S]*?<\\/footer>/gi, "");
+
+// Convert headings
+md = md.replace(/<h1[^>]*>(.*?)<\\/h1>/gi, "# $1\\n");
+md = md.replace(/<h2[^>]*>(.*?)<\\/h2>/gi, "## $1\\n");
+md = md.replace(/<h3[^>]*>(.*?)<\\/h3>/gi, "### $1\\n");
+md = md.replace(/<h4[^>]*>(.*?)<\\/h4>/gi, "#### $1\\n");
+
+// Convert code blocks
+md = md.replace(/<pre[^>]*><code[^>]*>(.*?)<\\/code><\\/pre>/gis, "\`\`\`\\n$1\\n\`\`\`\\n");
+md = md.replace(/<code[^>]*>(.*?)<\\/code>/gi, "\`$1\`");
+
+// Convert links
+md = md.replace(/<a[^>]*href="([^"]*)"[^>]*>(.*?)<\\/a>/gi, "[$2]($1)");
+
+// Convert lists
+md = md.replace(/<li[^>]*>(.*?)<\\/li>/gi, "- $1\\n");
+
+// Convert paragraphs
+md = md.replace(/<p[^>]*>(.*?)<\\/p>/gis, "$1\\n\\n");
+md = md.replace(/<br\\s*\\/?>/gi, "\\n");
+
+// Strip remaining tags
+md = md.replace(/<[^>]+>/g, "");
+
+// Decode entities
+md = md.replace(/&lt;/g, "<")
+  .replace(/&gt;/g, ">")
+  .replace(/&quot;/g, '"')
+  .replace(/&#39;/g, "'")
+  .replace(/&apos;/g, "'")
+  .replace(/&nbsp;/g, " ")
+  .replace(/&#(\\d+);/g, (_, n) => { const c = parseInt(n, 10); return c > 0 && c <= 0x10FFFF ? String.fromCodePoint(c) : ''; })
+  .replace(/&#x([0-9a-fA-F]+);/g, (_, h) => { const c = parseInt(h, 16); return c > 0 && c <= 0x10FFFF ? String.fromCodePoint(c) : ''; })
+  .replace(/&amp;/g, "&");
+
+// Clean whitespace
+md = md.replace(/\\n{3,}/g, "\\n\\n").trim();
+
+console.log(md);
+`;
+}
+//# sourceMappingURL=fetch-code.js.map

package/dist/util/fetch-code.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"fetch-code.js","sourceRoot":"","sources":["../../src/util/fetch-code.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AACH,MAAM,UAAU,cAAc,CAAC,GAAW,EAAE,UAA0B;IACrE,IAAI,UAAkB,CAAC;IACvB,IAAI,UAAU,EAAE,CAAC;QAChB,MAAM,SAAS,GAAG,IAAI,GAAG,CAAC,GAAG,CAAC,CAAC;QAC/B,MAAM,YAAY,GAAG,SAAS,CAAC,IAAI,CAAC;QACpC,wEAAwE;QACxE,sEAAsE;QACtE,2DAA2D;QAC3D,MAAM,aAAa,GAClB,UAAU,CAAC,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,UAAU,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,UAAU,GAAG,CAAC,CAAC,CAAC,UAAU,CAAC;QAC1F,SAAS,CAAC,QAAQ,GAAG,aAAa,CAAC;QACnC,UAAU,GAAG;cACD,IAAI,CAAC,SAAS,CAAC,SAAS,CAAC,QAAQ,EAAE,CAAC;qDACG,IAAI,CAAC,SAAS,CAAC,YAAY,CAAC,2BAA2B,CAAC;IAC5G,CAAC;SAAM,CAAC;QACP,UAAU,GAAG;cACD,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC;sDACqB,CAAC;IACtD,CAAC;IACD,OAAO,GAAG,UAAU;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAyDpB,CAAC;AACF,CAAC"}

package/dist/util/intent-filter.d.ts

@@ -0,0 +1,17 @@
+import type { Config } from "../config.js";
+import type { SessionTracker } from "../stats.js";
+import type { ContentStore } from "../store.js";
+interface IntentFilterDeps {
+    config: Config;
+    store: ContentStore;
+    tracker: SessionTracker;
+}
+/**
+ * Index large output and return a compact summary keyed to `intent`.
+ * For small output (<= config.intentSearchThreshold bytes), returns the
+ * original output unchanged so callers don't pay for trivial filtering.
+ */
+export declare function createIntentFilter(deps: IntentFilterDeps): (output: string, intent: string, sourceLabel: string) => string;
+export type ApplyIntentFilter = ReturnType<typeof createIntentFilter>;
+export {};
+//# sourceMappingURL=intent-filter.d.ts.map

package/dist/util/intent-filter.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"intent-filter.d.ts","sourceRoot":"","sources":["../../src/util/intent-filter.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,cAAc,CAAC;AAC3C,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAClD,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAGhD,UAAU,gBAAgB;IACzB,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,YAAY,CAAC;IACpB,OAAO,EAAE,cAAc,CAAC;CACxB;AAED;;;;GAIG;AACH,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,gBAAgB,IAGtB,QAAQ,MAAM,EAAE,QAAQ,MAAM,EAAE,aAAa,MAAM,KAAG,MAAM,CAoB9F;AAED,MAAM,MAAM,iBAAiB,GAAG,UAAU,CAAC,OAAO,kBAAkB,CAAC,CAAC"}

package/dist/util/intent-filter.js

@@ -0,0 +1,28 @@
+import { compactLabel } from "./label.js";
+/**
+ * Index large output and return a compact summary keyed to `intent`.
+ * For small output (<= config.intentSearchThreshold bytes), returns the
+ * original output unchanged so callers don't pay for trivial filtering.
+ */
+export function createIntentFilter(deps) {
+    const { config, store, tracker } = deps;
+    return function applyIntentFilter(output, intent, sourceLabel) {
+        if (Buffer.byteLength(output) <= config.intentSearchThreshold)
+            return output;
+        const indexed = store.index(output, sourceLabel);
+        tracker.trackIndexed(Buffer.byteLength(output));
+        const searchResults = store.search(intent, { limit: 3 });
+        const terms = store.getDistinctiveTerms(indexed.sourceId);
+        let filtered = `Indexed ${indexed.totalChunks} sections from ${sourceLabel}.\n`;
+        filtered += `${searchResults.results.length} sections matched "${intent}":\n\n`;
+        for (const hit of searchResults.results) {
+            filtered += `  - **${hit.title}**: ${hit.snippet.slice(0, 200)}\n`;
+        }
+        if (terms.length > 0 && config.compressionLevel !== "ultra") {
+            filtered += `\nSearchable terms: ${terms.join(", ")}\n`;
+        }
+        filtered += "\nUse search(queries: [...]) to retrieve full content of any section.";
+        return compactLabel(filtered, config.compressionLevel);
+    };
+}
+//# sourceMappingURL=intent-filter.js.map

package/dist/util/intent-filter.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"intent-filter.js","sourceRoot":"","sources":["../../src/util/intent-filter.ts"],"names":[],"mappings":"AAGA,OAAO,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAQ1C;;;;GAIG;AACH,MAAM,UAAU,kBAAkB,CAAC,IAAsB;IACxD,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,GAAG,IAAI,CAAC;IAExC,OAAO,SAAS,iBAAiB,CAAC,MAAc,EAAE,MAAc,EAAE,WAAmB;QACpF,IAAI,MAAM,CAAC,UAAU,CAAC,MAAM,CAAC,IAAI,MAAM,CAAC,qBAAqB;YAAE,OAAO,MAAM,CAAC;QAE7E,MAAM,OAAO,GAAG,KAAK,CAAC,KAAK,CAAC,MAAM,EAAE,WAAW,CAAC,CAAC;QACjD,OAAO,CAAC,YAAY,CAAC,MAAM,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC;QAEhD,MAAM,aAAa,GAAG,KAAK,CAAC,MAAM,CAAC,MAAM,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC,CAAC;QACzD,MAAM,KAAK,GAAG,KAAK,CAAC,mBAAmB,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;QAE1D,IAAI,QAAQ,GAAG,WAAW,OAAO,CAAC,WAAW,kBAAkB,WAAW,KAAK,CAAC;QAChF,QAAQ,IAAI,GAAG,aAAa,CAAC,OAAO,CAAC,MAAM,sBAAsB,MAAM,QAAQ,CAAC;QAChF,KAAK,MAAM,GAAG,IAAI,aAAa,CAAC,OAAO,EAAE,CAAC;YACzC,QAAQ,IAAI,SAAS,GAAG,CAAC,KAAK,OAAO,GAAG,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,IAAI,CAAC;QACpE,CAAC;QACD,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC,IAAI,MAAM,CAAC,gBAAgB,KAAK,OAAO,EAAE,CAAC;YAC7D,QAAQ,IAAI,uBAAuB,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC;QACzD,CAAC;QACD,QAAQ,IAAI,uEAAuE,CAAC;QACpF,OAAO,YAAY,CAAC,QAAQ,EAAE,MAAM,CAAC,gBAAgB,CAAC,CAAC;IACxD,CAAC,CAAC;AACH,CAAC"}

package/dist/util/label.d.ts

@@ -0,0 +1,4 @@
+import type { CompressionLevel } from "../config.js";
+/** Shorten labels based on compression level */
+export declare function compactLabel(normal: string, level: CompressionLevel): string;
+//# sourceMappingURL=label.d.ts.map

package/dist/util/label.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"label.d.ts","sourceRoot":"","sources":["../../src/util/label.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,cAAc,CAAC;AAErD,gDAAgD;AAChD,wBAAgB,YAAY,CAAC,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,gBAAgB,GAAG,MAAM,CAc5E"}

package/dist/util/label.js

@@ -0,0 +1,14 @@
+/** Shorten labels based on compression level */
+export function compactLabel(normal, level) {
+    if (level === "ultra") {
+        return normal
+            .replace(/\*\*/g, "")
+            .replace(/Use search\(queries: \[\.\.\.]\) to retrieve.*$/gm, "→ search() for more")
+            .replace(/Searchable terms: .+$/gm, "");
+    }
+    if (level === "compact") {
+        return normal.replace(/Use search\(queries: \[\.\.\.]\) to retrieve full content of any section\./, "→ search() for details");
+    }
+    return normal;
+}
+//# sourceMappingURL=label.js.map

package/dist/util/label.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"label.js","sourceRoot":"","sources":["../../src/util/label.ts"],"names":[],"mappings":"AAEA,gDAAgD;AAChD,MAAM,UAAU,YAAY,CAAC,MAAc,EAAE,KAAuB;IACnE,IAAI,KAAK,KAAK,OAAO,EAAE,CAAC;QACvB,OAAO,MAAM;aACX,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC;aACpB,OAAO,CAAC,mDAAmD,EAAE,qBAAqB,CAAC;aACnF,OAAO,CAAC,yBAAyB,EAAE,EAAE,CAAC,CAAC;IAC1C,CAAC;IACD,IAAI,KAAK,KAAK,SAAS,EAAE,CAAC;QACzB,OAAO,MAAM,CAAC,OAAO,CACpB,4EAA4E,EAC5E,wBAAwB,CACxB,CAAC;IACH,CAAC;IACD,OAAO,MAAM,CAAC;AACf,CAAC"}

package/dist/util/path.d.ts

@@ -0,0 +1,8 @@
+/**
+ * Returns true when `absPath` resolves inside (or equal to) `projectDir`.
+ * Uses realpathSync to defeat symlink-based escapes when the path exists,
+ * falling back to a string-prefix check for paths that don't exist yet
+ * (e.g. files about to be written).
+ */
+export declare function isWithinProject(absPath: string, projectDir: string): boolean;
+//# sourceMappingURL=path.d.ts.map

package/dist/util/path.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"path.d.ts","sourceRoot":"","sources":["../../src/util/path.ts"],"names":[],"mappings":"AAGA;;;;;GAKG;AACH,wBAAgB,eAAe,CAAC,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,GAAG,OAAO,CAU5E"}

package/dist/util/path.js

@@ -0,0 +1,21 @@
+import { realpathSync } from "node:fs";
+import { resolve } from "node:path";
+/**
+ * Returns true when `absPath` resolves inside (or equal to) `projectDir`.
+ * Uses realpathSync to defeat symlink-based escapes when the path exists,
+ * falling back to a string-prefix check for paths that don't exist yet
+ * (e.g. files about to be written).
+ */
+export function isWithinProject(absPath, projectDir) {
+    try {
+        const normalized = realpathSync(resolve(absPath));
+        const realProjectDir = realpathSync(projectDir);
+        return normalized === realProjectDir || normalized.startsWith(`${realProjectDir}/`);
+    }
+    catch {
+        const normalized = resolve(absPath);
+        const normalizedProject = resolve(projectDir);
+        return normalized === normalizedProject || normalized.startsWith(`${normalizedProject}/`);
+    }
+}
+//# sourceMappingURL=path.js.map

package/dist/util/path.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"path.js","sourceRoot":"","sources":["../../src/util/path.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AACvC,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAEpC;;;;;GAKG;AACH,MAAM,UAAU,eAAe,CAAC,OAAe,EAAE,UAAkB;IAClE,IAAI,CAAC;QACJ,MAAM,UAAU,GAAG,YAAY,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC;QAClD,MAAM,cAAc,GAAG,YAAY,CAAC,UAAU,CAAC,CAAC;QAChD,OAAO,UAAU,KAAK,cAAc,IAAI,UAAU,CAAC,UAAU,CAAC,GAAG,cAAc,GAAG,CAAC,CAAC;IACrF,CAAC;IAAC,MAAM,CAAC;QACR,MAAM,UAAU,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC;QACpC,MAAM,iBAAiB,GAAG,OAAO,CAAC,UAAU,CAAC,CAAC;QAC9C,OAAO,UAAU,KAAK,iBAAiB,IAAI,UAAU,CAAC,UAAU,CAAC,GAAG,iBAAiB,GAAG,CAAC,CAAC;IAC3F,CAAC;AACF,CAAC"}

package/dist/util/stream-compress.d.ts

@@ -0,0 +1,36 @@
+/**
+ * Line-by-line streaming compressor for long-running commands.
+ *
+ * Unlike the buffered pipeline (executor.ts / cli/filter.ts), this can emit
+ * compressed output before the child process exits. Necessary for
+ * `tail -f`, `cargo watch`, build commands with progressive output, etc.
+ *
+ * Trade-off: only stream-safe transformations are applied —
+ *   - ANSI stripping (per-line, no state)
+ *   - Progress/spinner line removal (per-line, no state)
+ *   - Adjacent-duplicate dedup (single-line lookback)
+ *
+ * Skipped because they need full output:
+ *   - applyCommandFilter (needs to detect summary/test markers globally)
+ *   - groupErrorLines (needs full set of error patterns)
+ *   - smartTruncate (needs final length)
+ */
+/**
+ * Stateful streaming compressor. Feed it chunks via `process(chunk)`, get
+ * compressed output back. Call `flush()` at end-of-stream to drain any
+ * buffered incomplete line + emit any pending dedup counter.
+ */
+export declare class StreamCompressor {
+    private buffer;
+    private prevLine;
+    private repeatCount;
+    /** Process a chunk; returns the filtered output ready to emit (may be empty). */
+    process(chunk: string): string;
+    /** Drain any remaining buffered line and emit final dedup counter. */
+    flush(): string;
+    /** Process the partial line in `buffer` (no trailing newline) and clear the buffer. */
+    private consumeBufferedLine;
+    /** Emit "(×N identical lines)" if a repeat counter is pending; reset state. */
+    private drainRepeatCounter;
+}
+//# sourceMappingURL=stream-compress.d.ts.map

package/dist/util/stream-compress.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"stream-compress.d.ts","sourceRoot":"","sources":["../../src/util/stream-compress.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAqBH;;;;GAIG;AACH,qBAAa,gBAAgB;IAC5B,OAAO,CAAC,MAAM,CAAM;IACpB,OAAO,CAAC,QAAQ,CAAuB;IACvC,OAAO,CAAC,WAAW,CAAK;IAExB,iFAAiF;IACjF,OAAO,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM;IA2B9B,sEAAsE;IACtE,KAAK,IAAI,MAAM;IASf,uFAAuF;IACvF,OAAO,CAAC,mBAAmB;IAa3B,+EAA+E;IAC/E,OAAO,CAAC,kBAAkB;CAK1B"}

package/dist/util/stream-compress.js

@@ -0,0 +1,104 @@
+/**
+ * Line-by-line streaming compressor for long-running commands.
+ *
+ * Unlike the buffered pipeline (executor.ts / cli/filter.ts), this can emit
+ * compressed output before the child process exits. Necessary for
+ * `tail -f`, `cargo watch`, build commands with progressive output, etc.
+ *
+ * Trade-off: only stream-safe transformations are applied —
+ *   - ANSI stripping (per-line, no state)
+ *   - Progress/spinner line removal (per-line, no state)
+ *   - Adjacent-duplicate dedup (single-line lookback)
+ *
+ * Skipped because they need full output:
+ *   - applyCommandFilter (needs to detect summary/test markers globally)
+ *   - groupErrorLines (needs full set of error patterns)
+ *   - smartTruncate (needs final length)
+ */
+// biome-ignore lint/suspicious/noControlCharactersInRegex: ANSI escape detection requires \x1b
+const ANSI_RE_G = /\x1b\[[0-9;]*[a-zA-Z]/g;
+const PROGRESS_BAR_RE = /^[\s\[│├└─═━▓░█▒▏▎▍▌▋▊▉\]>=#\-.\d%]+$/;
+const SPINNER_RE = /^[⠋⠙⠹⠸⠼⠴⠦⠧⠇⠏\-\\|/]\s/;
+const DOWNLOAD_RE = /(?:downloading|uploading|fetching|resolving)\s+[\d.]+\s*[kmg]?b/i;
+const SPEED_ETA_RE = /\d+\.?\d*\s*[kmg]?b\/s/i;
+const ETA_RE = /eta|remaining/i;
+function isProgressLine(line) {
+    const t = line.trim();
+    if (t === "")
+        return false; // keep empty lines (they delimit blocks)
+    if (PROGRESS_BAR_RE.test(t) && t.length > 3)
+        return true;
+    if (SPINNER_RE.test(t))
+        return true;
+    if (DOWNLOAD_RE.test(t))
+        return true;
+    if (SPEED_ETA_RE.test(t) && ETA_RE.test(t))
+        return true;
+    return false;
+}
+/**
+ * Stateful streaming compressor. Feed it chunks via `process(chunk)`, get
+ * compressed output back. Call `flush()` at end-of-stream to drain any
+ * buffered incomplete line + emit any pending dedup counter.
+ */
+export class StreamCompressor {
+    buffer = "";
+    prevLine = null;
+    repeatCount = 0;
+    /** Process a chunk; returns the filtered output ready to emit (may be empty). */
+    process(chunk) {
+        this.buffer += chunk;
+        const lastNl = this.buffer.lastIndexOf("\n");
+        if (lastNl < 0)
+            return ""; // no complete line yet
+        const complete = this.buffer.slice(0, lastNl);
+        this.buffer = this.buffer.slice(lastNl + 1);
+        let out = "";
+        for (const raw of complete.split("\n")) {
+            const filtered = raw.replace(ANSI_RE_G, "");
+            if (isProgressLine(filtered))
+                continue;
+            if (filtered === this.prevLine && filtered.trim() !== "") {
+                this.repeatCount++;
+                continue;
+            }
+            if (this.repeatCount > 1) {
+                out += `  ... (×${this.repeatCount} identical lines)\n`;
+            }
+            this.repeatCount = 0;
+            out += `${filtered}\n`;
+            this.prevLine = filtered;
+        }
+        return out;
+    }
+    /** Drain any remaining buffered line and emit final dedup counter. */
+    flush() {
+        let out = "";
+        if (this.buffer.length > 0) {
+            out += this.consumeBufferedLine();
+        }
+        out += this.drainRepeatCounter();
+        return out;
+    }
+    /** Process the partial line in `buffer` (no trailing newline) and clear the buffer. */
+    consumeBufferedLine() {
+        const filtered = this.buffer.replace(ANSI_RE_G, "");
+        this.buffer = "";
+        if (isProgressLine(filtered))
+            return "";
+        if (filtered === this.prevLine && filtered.trim() !== "") {
+            this.repeatCount++;
+            return "";
+        }
+        const counter = this.drainRepeatCounter();
+        this.prevLine = filtered;
+        return counter + filtered + (filtered.endsWith("\n") ? "" : "\n");
+    }
+    /** Emit "(×N identical lines)" if a repeat counter is pending; reset state. */
+    drainRepeatCounter() {
+        const out = this.repeatCount > 1 ? `  ... (×${this.repeatCount} identical lines)\n` : "";
+        this.repeatCount = 0;
+        return out;
+    }
+}
+//# sourceMappingURL=stream-compress.js.map

package/dist/util/stream-compress.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"stream-compress.js","sourceRoot":"","sources":["../../src/util/stream-compress.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAEH,+FAA+F;AAC/F,MAAM,SAAS,GAAG,wBAAwB,CAAC;AAE3C,MAAM,eAAe,GAAG,uCAAuC,CAAC;AAChE,MAAM,UAAU,GAAG,uBAAuB,CAAC;AAC3C,MAAM,WAAW,GAAG,kEAAkE,CAAC;AACvF,MAAM,YAAY,GAAG,yBAAyB,CAAC;AAC/C,MAAM,MAAM,GAAG,gBAAgB,CAAC;AAEhC,SAAS,cAAc,CAAC,IAAY;IACnC,MAAM,CAAC,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;IACtB,IAAI,CAAC,KAAK,EAAE;QAAE,OAAO,KAAK,CAAC,CAAC,yCAAyC;IACrE,IAAI,eAAe,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,MAAM,GAAG,CAAC;QAAE,OAAO,IAAI,CAAC;IACzD,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IACpC,IAAI,WAAW,CAAC,IAAI,CAAC,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IACrC,IAAI,YAAY,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IACxD,OAAO,KAAK,CAAC;AACd,CAAC;AAED;;;;GAIG;AACH,MAAM,OAAO,gBAAgB;IACpB,MAAM,GAAG,EAAE,CAAC;IACZ,QAAQ,GAAkB,IAAI,CAAC;IAC/B,WAAW,GAAG,CAAC,CAAC;IAExB,iFAAiF;IACjF,OAAO,CAAC,KAAa;QACpB,IAAI,CAAC,MAAM,IAAI,KAAK,CAAC;QACrB,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC;QAC7C,IAAI,MAAM,GAAG,CAAC;YAAE,OAAO,EAAE,CAAC,CAAC,uBAAuB;QAClD,MAAM,QAAQ,GAAG,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC;QAC9C,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QAE5C,IAAI,GAAG,GAAG,EAAE,CAAC;QACb,KAAK,MAAM,GAAG,IAAI,QAAQ,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC;YACxC,MAAM,QAAQ,GAAG,GAAG,CAAC,OAAO,CAAC,SAAS,EAAE,EAAE,CAAC,CAAC;YAC5C,IAAI,cAAc,CAAC,QAAQ,CAAC;gBAAE,SAAS;YAEvC,IAAI,QAAQ,KAAK,IAAI,CAAC,QAAQ,IAAI,QAAQ,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC;gBAC1D,IAAI,CAAC,WAAW,EAAE,CAAC;gBACnB,SAAS;YACV,CAAC;YAED,IAAI,IAAI,CAAC,WAAW,GAAG,CAAC,EAAE,CAAC;gBAC1B,GAAG,IAAI,WAAW,IAAI,CAAC,WAAW,qBAAqB,CAAC;YACzD,CAAC;YACD,IAAI,CAAC,WAAW,GAAG,CAAC,CAAC;YACrB,GAAG,IAAI,GAAG,QAAQ,IAAI,CAAC;YACvB,IAAI,CAAC,QAAQ,GAAG,QAAQ,CAAC;QAC1B,CAAC;QACD,OAAO,GAAG,CAAC;IACZ,CAAC;IAED,sEAAsE;IACtE,KAAK;QACJ,IAAI,GAAG,GAAG,EAAE,CAAC;QACb,IAAI,IAAI,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC5B,GAAG,IAAI,IAAI,CAAC,mBAAmB,EAAE,CAAC;QACnC,CAAC;QACD,GAAG,IAAI,IAAI,CAAC,kBAAkB,EAAE,CAAC;QACjC,OAAO,GAAG,CAAC;IACZ,CAAC;IAED,uFAAuF;IAC/E,mBAAmB;QAC1B,MAAM,QAAQ,GAAG,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,SAAS,EAAE,EAAE,CAAC,CAAC;QACpD,IAAI,CAAC,MAAM,GAAG,EAAE,CAAC;QACjB,IAAI,cAAc,CAAC,QAAQ,CAAC;YAAE,OAAO,EAAE,CAAC;QACxC,IAAI,QAAQ,KAAK,IAAI,CAAC,QAAQ,IAAI,QAAQ,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC;YAC1D,IAAI,CAAC,WAAW,EAAE,CAAC;YACnB,OAAO,EAAE,CAAC;QACX,CAAC;QACD,MAAM,OAAO,GAAG,IAAI,CAAC,kBAAkB,EAAE,CAAC;QAC1C,IAAI,CAAC,QAAQ,GAAG,QAAQ,CAAC;QACzB,OAAO,OAAO,GAAG,QAAQ,GAAG,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;IACnE,CAAC;IAED,+EAA+E;IACvE,kBAAkB;QACzB,MAAM,GAAG,GAAG,IAAI,CAAC,WAAW,GAAG,CAAC,CAAC,CAAC,CAAC,WAAW,IAAI,CAAC,WAAW,qBAAqB,CAAC,CAAC,CAAC,EAAE,CAAC;QACzF,IAAI,CAAC,WAAW,GAAG,CAAC,CAAC;QACrB,OAAO,GAAG,CAAC;IACZ,CAAC;CACD"}

package/dist/util/version.d.ts

@@ -0,0 +1,2 @@
+export declare function getVersion(fallback?: string): string;
+//# sourceMappingURL=version.d.ts.map

package/dist/util/version.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"version.d.ts","sourceRoot":"","sources":["../../src/util/version.ts"],"names":[],"mappings":"AAMA,wBAAgB,UAAU,CAAC,QAAQ,SAAU,GAAG,MAAM,CAQrD"}

package/dist/util/version.js

@@ -0,0 +1,15 @@
+import { readFileSync } from "node:fs";
+import { dirname, resolve } from "node:path";
+import { fileURLToPath } from "node:url";
+const __dirname = dirname(fileURLToPath(import.meta.url));
+export function getVersion(fallback = "1.0.0") {
+    try {
+        const pkgPath = resolve(__dirname, "..", "..", "package.json");
+        const pkg = JSON.parse(readFileSync(pkgPath, "utf-8"));
+        return pkg.version ?? fallback;
+    }
+    catch {
+        return fallback;
+    }
+}
+//# sourceMappingURL=version.js.map

package/dist/util/version.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"version.js","sourceRoot":"","sources":["../../src/util/version.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AACvC,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAC7C,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AAEzC,MAAM,SAAS,GAAG,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;AAE1D,MAAM,UAAU,UAAU,CAAC,QAAQ,GAAG,OAAO;IAC5C,IAAI,CAAC;QACJ,MAAM,OAAO,GAAG,OAAO,CAAC,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,cAAc,CAAC,CAAC;QAC/D,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC;QACvD,OAAO,GAAG,CAAC,OAAO,IAAI,QAAQ,CAAC;IAChC,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,QAAQ,CAAC;IACjB,CAAC;AACF,CAAC"}

package/docs/agentic-benchmark.md

@@ -0,0 +1,110 @@
+# Agentic Benchmark Plan
+
+This benchmark measures context-compress in real agent sessions, not synthetic command output alone.
+
+The claim to test:
+
+> Large tool output should stay searchable outside the conversation, while the agent still solves the same task with less context pressure.
+
+## Why This Exists
+
+`docs/token-reduction-report.md` measures byte and token reduction for common operations. That is necessary, but it does not fully answer whether an agent remains effective across a real coding task.
+
+This benchmark adds the missing layer: run the same task with and without context-compress, isolate each arm, and compare context usage, task success, cost, and time.
+
+## Arms
+
+| Arm | Setup | Purpose |
+| --- | --- | --- |
+| `baseline` | No context-compress MCP, no hook | Measures normal agent behavior. |
+| `mcp-only` | MCP server registered, no PreToolUse hook | Measures explicit tool adoption. |
+| `hook-balanced` | MCP plus PreToolUse hook, `CONTEXT_COMPRESS_MODE=balanced` | Default recommended setup. |
+| `hook-aggressive` | MCP plus PreToolUse hook, `CONTEXT_COMPRESS_MODE=aggressive` | Maximum compression trade-off. |
+
+Each arm must run in a fresh workspace with isolated agent settings. Do not allow global plugins, global MCP servers, or previous conversation state to leak into the run.
+
+## Task Set
+
+Use tasks that naturally produce large outputs:
+
+1. Diagnose a failing test suite and patch the root cause.
+2. Review a multi-commit diff and summarize risky changes.
+3. Inspect a large API response and implement one missing field mapping.
+4. Analyze a generated Playwright snapshot and fix one selector bug.
+5. Audit dependency output and identify one vulnerable or outdated package.
+6. Search a large log file and explain the first recurring failure.
+
+Pin every input repository and fixture by commit hash. Preserve every run directory so metrics can be recomputed.
+
+## Metrics
+
+| Metric | How to collect |
+| --- | --- |
+| Context bytes returned by tools | Sum raw tool payloads in agent logs. |
+| Compressed bytes returned | Sum context-compress tool responses. |
+| Indexed bytes | Use `stats` output and session DB stats. |
+| Task success | Deterministic test, assertion, or scorer per task. |
+| Cost/time | Agent runner JSON output when available. |
+| Follow-up retrieval quality | Count whether the final answer cites indexed/search results when needed. |
+
+Report raw numbers and relative deltas. Do not only report the best percentage.
+
+## Isolation Rules
+
+- Use a new temp workspace for every `(task, arm, run)` cell.
+- Disable user/global plugin sources for the baseline arm.
+- Install exactly the intended plugin or MCP config for non-baseline arms.
+- Clear persistent context-compress DBs between runs unless the task explicitly tests persistence.
+- Keep model, prompt, timeout, and working tree identical across arms.
+- Record the exact agent version, model, OS, Node version, and context-compress version.
+
+## Safety Checks
+
+Compression must not hide important failures. Every task needs one deterministic scorer:
+
+- tests pass after the agent patch,
+- expected files changed and unrelated files did not,
+- security-relevant details are still retrievable with `search`,
+- final answer includes the actual root cause, not just a compressed summary.
+
+If an arm uses fewer tokens but fails the scorer, mark it as a failure, not a win.
+
+## Reporting Template
+
+```md
+# Agentic benchmark: context-compress on real coding tasks
+
+Date:
+Agent:
+Model:
+context-compress:
+Repo/fixture commits:
+
+## Summary
+
+| Arm | Success | Tool bytes in context | Indexed bytes | Cost | Time |
+| --- | ---: | ---: | ---: | ---: | ---: |
+
+## Per-task Results
+
+| Task | Arm | Success | Tool bytes | Indexed bytes | Notes |
+| --- | --- | ---: | ---: | ---: | --- |
+
+## Failures And Limits
+
+- What failed:
+- What this benchmark does not prove:
+- Known nondeterminism:
+```
+
+## Reproduce
+
+Until this harness is automated, run the benchmark manually with:
+
+```bash
+npm run build
+context-compress setup --auto
+CONTEXT_COMPRESS_MODE=balanced context-compress doctor
+```
+
+Then run each task in isolated agent settings and attach the resulting logs plus `context-compress stats` output to the benchmark result.