klyro 0.1.40 → 0.1.42

package/dist/agent/runtime.d.ts

@@ -191,6 +191,11 @@ export interface RunResult {
 }
 /** Convert a registry of tools into ToolDefinitions for the provider. */
 export declare function toolDefinitions(registry: ToolRegistry): ToolDefinition[];
+/** Estimate USD cost of a usage block given the model name. */
+export declare function estimateCost(model: string, usage: {
+    input: number;
+    output: number;
+}): number;
 /** Run the autonomous loop. */
 export declare function run(opts: RunOptions, deps: RuntimeDeps): Promise<RunResult>;
 export declare function defaultSystemPrompt(ctx: {

package/dist/agent/runtime.js

@@ -35,6 +35,36 @@ export function toolDefinitions(registry) {
         inputSchema: t.function.parameters,
     }));
 }
+// BUG-005: Model-aware cost estimation with sensible defaults.
+// Rates are per-1K tokens (input / output). Local models are $0.
+const MODEL_RATES = [
+    { test: (m) => /gpt-4/i.test(m), input: 0.003, output: 0.015 },
+    { test: (m) => /gpt-3\.5/i.test(m), input: 0.0005, output: 0.0015 },
+    { test: (m) => /claude|anthropic/i.test(m), input: 0.003, output: 0.015 },
+    { test: (m) => /gemini/i.test(m), input: 0.00075, output: 0.003 },
+    { test: (m) => /o1/i.test(m), input: 0.015, output: 0.06 },
+];
+/** Estimate USD cost of a usage block given the model name. */
+export function estimateCost(model, usage) {
+    const match = MODEL_RATES.find((r) => r.test(model));
+    const { input: inRate, output: outRate } = match ?? { input: 0.003, output: 0.015 };
+    return (usage.input / 1000) * inRate + (usage.output / 1000) * outRate;
+}
+// PERF-002: Memoized token counting cache.
+let tokenCache = {
+    lastRef: null,
+    lastSystem: undefined,
+    lastCount: 0,
+};
+/** Memoized totalTokens � only recomputes when transcript ref or system changes. */
+function cachedTotalTokens(system, messages) {
+    if (tokenCache.lastRef === messages && tokenCache.lastSystem === system) {
+        return tokenCache.lastCount;
+    }
+    const count = totalTokens(system, messages);
+    tokenCache = { lastRef: messages, lastSystem: system, lastCount: count };
+    return count;
+}
 /** Run the autonomous loop. */
 export async function run(opts, deps) {
     const maxSteps = opts.maxTurns ?? opts.maxSteps ?? DEFAULT_MAX_STEPS;
@@ -143,12 +173,10 @@ export async function run(opts, deps) {
     // 5.2 — stuck detection state
     const callHistory = [];
     const fileEditCounts = new Map();
-    let stuckCount = 0;
-    let lastSignal;
     outer: while (steps < maxSteps) {
         // 5.1 limits: max-cost, max-time
         if (maxCost !== undefined) {
-            const cost = (usage.input / 1000) * 0.003 + (usage.output / 1000) * 0.015;
+            const cost = estimateCost(opts.model, usage);
             if (cost >= maxCost) {
                 setPhase('limit');
                 await closeTracer();
@@ -189,10 +217,11 @@ export async function run(opts, deps) {
         const BUDGET = { total: 120_000, reservedOutput: 4000 };
         let reqMessages = transcript;
         let reqSystem = systemPrompt;
-        if (totalTokens(systemPrompt, transcript) > BUDGET.total) {
+        if (cachedTotalTokens(systemPrompt, transcript) > BUDGET.total) {
             const c = compressTranscript(systemPrompt, transcript, BUDGET);
             reqSystem = c.system;
             reqMessages = c.messages;
+            tokenCache = { lastRef: null, lastSystem: undefined, lastCount: 0 };
             if (c.dropped > 0)
                 emitKlyro({ type: 'context.compacted', ts: Date.now(), sessionId: sessionId ?? 'ephemeral', dropped: c.dropped });
         }
@@ -279,6 +308,8 @@ export async function run(opts, deps) {
         const assistantMsg = { role: 'assistant', content: assistantContent };
         transcript.push(assistantMsg);
         await checkpoint(assistantMsg);
+        // Invalidate token cache since transcript changed
+        tokenCache = { lastRef: null, lastSystem: undefined, lastCount: 0 };
         // No tool calls → potential completion (Level 8 verify gate)
         if (opts.signal?.aborted) {
             finalText = textBuf;
@@ -363,7 +394,7 @@ export async function run(opts, deps) {
                 }
                 // 6.4 — classify
                 const baseline = await getBaseline(opts.cwd, verifyCmd);
-                const isFlaky = await rerunOnce(opts.cwd, cmdToRun, 30_000);
+                const isFlaky = await rerunOnce(opts.cwd, cmdToRun, opts.verify?.timeoutMs ?? 45_000);
                 const cls = classifyFailure({ failure: vResult.failure, stdout: vResult.stdout, stderr: vResult.stderr }, baseline, isFlaky);
                 if (cls === 'flaky') {
                     // rerun succeeded on second try — treat as flaky, don't count as repair
@@ -596,16 +627,15 @@ export async function run(opts, deps) {
             }
         };
         // 3.5 — parallel if all concurrencySafe, sequential otherwise
-        // For parallel, execute concurrently but commit transcript in original call order to preserve determinism
+        // BUG-002: preserve call order — run sequentially even when allSafe to avoid out-of-order transcript
+        // Parallel execution previously pushed tool_results out of order via Promise.all
         if (allSafe) {
-            toolCallCount += finalizedCalls.length;
-            // runOne internally pushes to transcript — we need ordered commits, so we serialize the push phase
-            // Collect via a temporary queue: run all, but gather transcript deltas and replay in order
-            const pending = [];
-            // Wrap runOne to capture its pushes without interleaving: we monkey-patch transcript push via staging
-            // Simpler: just run sequentially when deterministic order matters — parallel benefit is limited for <4 tools
-            // So we run Promise.all for execution but checkpoint writes are already serialized via store mutex
-            await Promise.all(finalizedCalls.map((c) => runOne(c)));
+            for (const call of finalizedCalls) {
+                toolCallCount++;
+                await runOne(call);
+                if (opts.signal?.aborted)
+                    break;
+            }
         }
         else {
             for (const call of finalizedCalls) {

package/dist/tools/fs/apply-patch.js

@@ -6,7 +6,8 @@ import * as path from 'node:path';
 import { z } from 'zod';
 import { defineTool } from '../types.js';
 import { resolveAndFollowSymlinks } from '../../policy/path-guard.js';
-import { safe } from '../normalize.js';
+import { safe, TOOL_ERROR_CODES } from '../normalize.js';
+import { wasRead } from './read-history.js';
 const InputSchema = z.object({
     patch: z.string().min(1).describe('Unified diff patch text'),
 });
@@ -29,6 +30,18 @@ export const applyPatchTool = defineTool({
                     // Flush previous
                     if (currentFile && fileContent !== null) {
                         const { resolved } = await resolveAndFollowSymlinks(ctx.cwd, currentFile);
+                        if (fileContent === null)
+                            throw Object.assign(new Error('fileContent is null — cannot write'), { code: TOOL_ERROR_CODES.INTERNAL });
+                        try {
+                            const existing = await fs.readFile(resolved, 'utf-8');
+                            if (!wasRead(currentFile) && existing.length > 200) {
+                                throw Object.assign(new Error('File was not read this session and is >200 bytes — re-read before writing (POLICY_DENIED)'), { code: 'POLICY_DENIED' });
+                            }
+                        }
+                        catch (e) {
+                            if (e && typeof e === 'object' && e.code === 'POLICY_DENIED')
+                                throw e;
+                        }
                         await fs.mkdir(path.dirname(resolved), { recursive: true });
                         await fs.writeFile(resolved, fileContent, 'utf-8');
                         patchedFiles.push(currentFile);
@@ -48,6 +61,18 @@ export const applyPatchTool = defineTool({
                 if (line.startsWith('*** Add File:')) {
                     if (currentFile && fileContent !== null) {
                         const { resolved } = await resolveAndFollowSymlinks(ctx.cwd, currentFile);
+                        if (fileContent === null)
+                            throw Object.assign(new Error('fileContent is null — cannot write'), { code: TOOL_ERROR_CODES.INTERNAL });
+                        try {
+                            const existing = await fs.readFile(resolved, 'utf-8');
+                            if (!wasRead(currentFile) && existing.length > 200) {
+                                throw Object.assign(new Error('File was not read this session and is >200 bytes — re-read before writing (POLICY_DENIED)'), { code: 'POLICY_DENIED' });
+                            }
+                        }
+                        catch (e) {
+                            if (e && typeof e === 'object' && e.code === 'POLICY_DENIED')
+                                throw e;
+                        }
                         await fs.mkdir(path.dirname(resolved), { recursive: true });
                         await fs.writeFile(resolved, fileContent, 'utf-8');
                         patchedFiles.push(currentFile);
@@ -67,6 +92,18 @@ export const applyPatchTool = defineTool({
             }
             if (currentFile && fileContent !== null) {
                 const { resolved } = await resolveAndFollowSymlinks(ctx.cwd, currentFile);
+                if (fileContent === null)
+                    throw Object.assign(new Error('fileContent is null — cannot write'), { code: TOOL_ERROR_CODES.INTERNAL });
+                try {
+                    const existing = await fs.readFile(resolved, 'utf-8');
+                    if (!wasRead(currentFile) && existing.length > 200) {
+                        throw Object.assign(new Error('File was not read this session and is >200 bytes — re-read before writing (POLICY_DENIED)'), { code: 'POLICY_DENIED' });
+                    }
+                }
+                catch (e) {
+                    if (e && typeof e === 'object' && e.code === 'POLICY_DENIED')
+                        throw e;
+                }
                 await fs.mkdir(path.dirname(resolved), { recursive: true });
                 await fs.writeFile(resolved, fileContent, 'utf-8');
                 patchedFiles.push(currentFile);

package/dist/tools/shell/shell-exec.js

@@ -48,10 +48,21 @@ function filteredEnv(extra) {
             out[k] = v;
         }
     }
-    // Always allow basic
+    // Merge user extra AFTER filtering — but only allowed prefixes, block injection vectors
+    if (extra) {
+        for (const [k, v] of Object.entries(extra)) {
+            if (k === 'NODE_OPTIONS' || k === 'LD_PRELOAD' || k === 'LD_LIBRARY_PATH')
+                continue;
+            if (k.includes('SECRET') || k.includes('TOKEN') || k === 'ANTHROPIC_API_KEY' || k === 'OPENAI_API_KEY')
+                continue;
+            if (ALLOWED_ENV_PREFIXES.some((p) => k.startsWith(p)) || k === 'PATH' || k === 'PWD' || k === 'TMPDIR' || k === 'TEMP') {
+                out[k] = v;
+            }
+        }
+    }
     out.PATH = process.env.PATH;
-    if (extra)
-        Object.assign(out, extra);
+    if (out.NODE_OPTIONS)
+        delete out.NODE_OPTIONS;
     return out;
 }
 // Interactive command detection (3.3)

package/dist/tui/app.d.ts

@@ -1,6 +1,6 @@
 /**
- * Klyro TUI — opencode-clean — no clumsy words, correct wrap, markdown, scroll
- * Header 3 rows, guide │ at col2, ● Klyro accent, prose wrapped at word boundaries
+ * Klyro TUI - opencode-clean - no clumsy words, correct wrap, markdown, scroll
+ * Header 3 rows, guide │ at col2, ● Klyro accent, prose wrapped at word boundaries
  */
 import React from 'react';
 import type { StatusSnapshot } from './status.js';

package/dist/tui/app.js

@@ -1,9 +1,9 @@
 import { jsxs as _jsxs, jsx as _jsx } from "react/jsx-runtime";
 /**
- * Klyro TUI — opencode-clean — no clumsy words, correct wrap, markdown, scroll
- * Header 3 rows, guide │ at col2, ● Klyro accent, prose wrapped at word boundaries
+ * Klyro TUI - opencode-clean - no clumsy words, correct wrap, markdown, scroll
+ * Header 3 rows, guide │ at col2, ● Klyro accent, prose wrapped at word boundaries
  */
-import { useState, useEffect, useRef, useCallback } from 'react';
+import { useState, useEffect, useRef, useCallback, useMemo } from 'react';
 import { Box, Text, useInput, useStdout } from 'ink';
 import { TuiApprovalBridge } from './approval.js';
 import { parse as parseSlash } from '../cli/slash/parser.js';
@@ -19,7 +19,7 @@ function Header({ cwd, model, version, width }) {
         return '';
     } })();
     const showLinks = width >= 120;
-    return (_jsxs(Box, { flexDirection: "column", marginBottom: 1, children: [_jsxs(Box, { justifyContent: "space-between", children: [_jsxs(Text, { bold: true, color: tokens.colors.accent, children: ["KLYRO  v", version] }), showLinks ? _jsx(Text, { color: tokens.colors.dim, children: "\u00E2\u201D\u201A  /help   /config   /clear   /exit" }) : null] }), _jsxs(Text, { color: tokens.colors.dim, children: [model, "[200k]  \u00C2\u00B7  API Usage Billing"] }), _jsxs(Text, { color: tokens.colors.dim, children: [cwd, branch ? `  ·  ${branch}` : ''] })] }));
+    return (_jsxs(Box, { flexDirection: "column", marginBottom: 1, children: [_jsxs(Box, { justifyContent: "space-between", children: [_jsxs(Text, { bold: true, color: tokens.colors.accent, children: ["KLYRO  v", version] }), showLinks ? _jsx(Text, { color: tokens.colors.dim, children: "\u2502  /help   /config   /clear   /exit" }) : null] }), _jsxs(Text, { color: tokens.colors.dim, children: [model, "[200k]  \u00B7  API Usage Billing"] }), _jsxs(Text, { color: tokens.colors.dim, children: [cwd, branch ? `  ·  ${branch}` : ''] })] }));
 }
 function verbForTool(name) {
     if (name === 'read_file')
@@ -71,7 +71,7 @@ function groupTools(items) {
     flush();
     return out;
 }
-// Simple markdown: **bold** → bold, keep lists/tables, wrap at word boundaries
+// Simple markdown: **bold** †’ bold, keep lists/tables, wrap at word boundaries
 function MarkdownText({ text, dim, width }) {
     // Split by **bold** segments
     const parts = [];
@@ -89,9 +89,82 @@ function MarkdownText({ text, dim, width }) {
         parts.push(_jsx(Text, { color: dim ? tokens.colors.dim : undefined, wrap: "wrap", children: text.slice(last) }, `t-${idx++}`));
     if (parts.length === 0)
         return _jsx(Text, { color: dim ? tokens.colors.dim : undefined, wrap: "wrap", children: text });
-    // Render as single line with bold segments — Ink will wrap the parent Box
+    // Render as single line with bold segments — Ink will wrap the parent Box
     return _jsx(Text, { wrap: "wrap", children: parts });
 }
+// Chat scroll state: scrollOffset, pinned (user scrolled away from bottom),
+// pendingNew (rows arrived while pinned), and a commands bag for key handlers.
+// The `tick` prop is a monotonic value that increments on every content mutation,
+// including in-place text growth during streaming (appendDelta mutates by index,
+// so transcript.length does not change on a delta — the effect must fire anyway).
+function useChatScroll(opts) {
+    const { totalRows, viewportH, messageBoundaries, tick } = opts;
+    const [scrollOffset, setScrollOffset] = useState(0);
+    const [pinned, setPinned] = useState(false);
+    const [pendingNew, setPendingNew] = useState(0);
+    const pinnedRef = useRef(false);
+    const maxOffset = Math.max(0, totalRows - viewportH);
+    // 1-line tolerance: maxOffset can shift by 1 during streaming and leave us
+    // at maxOffset - 1, which would otherwise be "not at bottom". The +1 tolerance
+    // keeps follow-tail engaged through that off-by-one.
+    const isAtBottom = scrollOffset + 1 >= maxOffset;
+    const recomputePinned = useCallback((next) => {
+        const atBottom = next + 1 >= maxOffset;
+        pinnedRef.current = !atBottom;
+        setPinned(!atBottom);
+        if (atBottom)
+            setPendingNew(0);
+    }, [maxOffset]);
+    // Watch `tick` — fires on every content change (add, remove, in-place delta).
+    const lastTickRef = useRef(tick);
+    const lastMaxOffsetRef = useRef(maxOffset);
+    const firstEffectRef = useRef(true);
+    useEffect(() => {
+        if (firstEffectRef.current) {
+            // Initial mount: if there's content, follow the tail (preserves the
+            // pre-refactor behavior where scrollOffset was 0 only on empty state).
+            firstEffectRef.current = false;
+            lastTickRef.current = tick;
+            lastMaxOffsetRef.current = maxOffset;
+            if (maxOffset > 0) {
+                setScrollOffset(maxOffset);
+            }
+            return;
+        }
+        if (tick === lastTickRef.current)
+            return;
+        lastTickRef.current = tick;
+        const grew = maxOffset - lastMaxOffsetRef.current;
+        lastMaxOffsetRef.current = maxOffset;
+        if (pinnedRef.current) {
+            if (grew > 0)
+                setPendingNew((p) => p + grew);
+        }
+        else {
+            // FollowTail: snap to the new bottom.
+            setScrollOffset(maxOffset);
+        }
+    }, [tick, maxOffset]);
+    const commands = {
+        lineUp: () => { const next = Math.max(0, scrollOffset - 1); setScrollOffset(next); recomputePinned(next); },
+        lineDown: () => { const next = Math.min(maxOffset, scrollOffset + 1); setScrollOffset(next); recomputePinned(next); },
+        pageUp: () => {
+            const prev = [...messageBoundaries].reverse().find((b) => b < scrollOffset);
+            const next = prev ?? Math.max(0, scrollOffset - viewportH);
+            setScrollOffset(next);
+            recomputePinned(next);
+        },
+        pageDown: () => {
+            const nxt = messageBoundaries.find((b) => b > scrollOffset);
+            const next = nxt ?? Math.min(maxOffset, scrollOffset + viewportH);
+            setScrollOffset(next);
+            recomputePinned(next);
+        },
+        jumpTop: () => { setScrollOffset(0); recomputePinned(0); },
+        jumpBottom: () => { setScrollOffset(maxOffset); recomputePinned(maxOffset); },
+    };
+    return { scrollOffset, setScrollOffset, pinned, pendingNew, isAtBottom, maxOffset, commands };
+}
 export function App(props) {
     const { stdout } = useStdout();
     const [transcript, setTranscript] = useState(props.initialTranscript ?? []);
@@ -103,7 +176,6 @@ export function App(props) {
     const [elapsed, setElapsed] = useState(0);
     const [queuedInputs, setQueuedInputs] = useState([]);
     const [expandedGroups, setExpandedGroups] = useState(new Set());
-    const [scrollOffset, setScrollOffset] = useState(0);
     const streamingIdRef = useRef(null);
     const width = stdout?.columns ?? 100;
     const height = stdout?.rows ?? 30;
@@ -111,8 +183,18 @@ export function App(props) {
     const grouped = groupTools(transcript);
     const viewportH = Math.max(5, height - 10);
     const totalRows = grouped.length + (plan.length > 0 ? 1 : 0) + 2;
-    const maxOffset = Math.max(0, totalRows - viewportH);
-    const isAtBottom = scrollOffset >= maxOffset;
+    const messageBoundaries = useMemo(() => grouped.map((_, i) => i), [grouped]);
+    // Monotonic tick: increments on every render, so the scroll hook fires
+    // for every content mutation — including in-place text deltas.
+    const tickRef = useRef(0);
+    useEffect(() => { tickRef.current += 1; });
+    const scroll = useChatScroll({
+        totalRows,
+        viewportH,
+        messageBoundaries,
+        tick: tickRef.current,
+    });
+    const { scrollOffset, isAtBottom, maxOffset, pinned, pendingNew, commands } = scroll;
     const trackH = viewportH;
     const thumbPos = maxOffset === 0 ? 0 : Math.round((scrollOffset / maxOffset) * (trackH - 1));
     const visibleGrouped = isFullscreen ? grouped.slice(scrollOffset, scrollOffset + viewportH) : grouped;
@@ -132,8 +214,6 @@ export function App(props) {
     }, [queuedInputs, status.status, awaitingApproval]);
     useEffect(() => { if (status.status !== 'running')
         return; const start = Date.now() - elapsed; const t = setInterval(() => setElapsed(Date.now() - start), 1000); return () => clearInterval(t); }, [status.status, elapsed]);
-    useEffect(() => { if (isAtBottom)
-        setScrollOffset(maxOffset); }, [transcript.length, plan.length, maxOffset, isAtBottom]);
     const append = useCallback((item) => { if (item.kind !== 'text' || item.role !== 'assistant')
         streamingIdRef.current = null; setTranscript((prev) => [...prev, item]); }, []);
     const appendDelta = useCallback((text) => {
@@ -160,28 +240,37 @@ export function App(props) {
         n.delete(id);
     else
         n.add(id); return n; });
-    const scrollUp = (n = 3) => setScrollOffset((p) => Math.max(0, p - n));
-    const scrollDown = (n = 3) => setScrollOffset((p) => Math.min(maxOffset, p + n));
     useInput((inputStr, key) => {
         if (key.escape && queuedInputs.length > 0) {
             setQueuedInputs((prev) => prev.slice(1));
             return;
         }
-        if (key.pageUp || (key.ctrl && inputStr === 'u')) {
-            scrollUp(5);
-            return;
-        }
-        if (key.pageDown || (key.ctrl && inputStr === 'd')) {
-            scrollDown(5);
-            return;
-        }
-        if (key.upArrow && (key.shift || key.ctrl)) {
-            scrollUp(1);
-            return;
-        }
-        if (key.downArrow && (key.shift || key.ctrl)) {
-            scrollDown(1);
-            return;
+        // Scroll keys (work in any mode, including while running).
+        if (isFullscreen && maxOffset > 0) {
+            if (key.home) {
+                commands.jumpTop();
+                return;
+            }
+            if (key.end) {
+                commands.jumpBottom();
+                return;
+            }
+            if (key.pageUp || (key.ctrl && inputStr === 'u')) {
+                commands.pageUp();
+                return;
+            }
+            if (key.pageDown || (key.ctrl && inputStr === 'd')) {
+                commands.pageDown();
+                return;
+            }
+            if (key.upArrow && (key.shift || key.ctrl)) {
+                commands.lineUp();
+                return;
+            }
+            if (key.downArrow && (key.shift || key.ctrl)) {
+                commands.lineDown();
+                return;
+            }
         }
         if (awaitingApproval)
             return;
@@ -241,9 +330,9 @@ export function App(props) {
     const cost = (status.usageInput / 1000 * 0.003 + status.usageOutput / 1000 * 0.015);
     const totalTokens = status.usageInput + status.usageOutput;
     const ctxPct = totalTokens > 0 ? Math.round((totalTokens / 120_000) * 100) : 0;
-    const baseHints = status.status === 'running' ? 'ctrl+c to stop  ·  enter to queue  ·  ctrl+o expand' : transcript.length === 0 ? 'shift+tab to cycle  ·  ↑↓ for history  ·  / for commands' : 'enter to send  ·  shift+enter newline  ·  @ to attach';
-    const hints = maxOffset > 0 && isFullscreen ? `${baseHints}  ·  PgUp/Dn scroll` : baseHints;
-    return (_jsxs(Box, { flexDirection: "column", width: width, height: isFullscreen ? height - 1 : undefined, children: [_jsx(Header, { cwd: props.cwd, model: status.model, version: ver, width: width }), _jsxs(Box, { flexDirection: "row", flexGrow: isFullscreen ? 1 : 0, overflow: isFullscreen ? 'hidden' : undefined, children: [_jsxs(Box, { flexDirection: "column", flexGrow: 1, overflow: isFullscreen ? 'hidden' : undefined, paddingX: 0, children: [grouped.length === 0 ? (_jsx(Text, { color: tokens.colors.dim, children: "Message Klyro\u00E2\u20AC\u00A6" })) : visibleGrouped.map((item) => {
+    const baseHints = status.status === 'running' ? 'ctrl+c to stop  ·  enter to queue  ·  ctrl+o expand' : transcript.length === 0 ? 'shift+tab to cycle  ·  †‘†“ for history  ·  / for commands' : 'enter to send  ·  shift+enter newline  ·  @ to attach';
+    const hints = maxOffset > 0 && isFullscreen ? `${baseHints}  ·  PgUp/Dn scroll` : baseHints;
+    return (_jsxs(Box, { flexDirection: "column", width: width, height: isFullscreen ? height - 1 : undefined, children: [_jsx(Header, { cwd: props.cwd, model: status.model, version: ver, width: width }), _jsxs(Box, { flexDirection: "row", flexGrow: isFullscreen ? 1 : 0, overflow: isFullscreen ? 'hidden' : undefined, children: [_jsxs(Box, { flexDirection: "column", flexGrow: 1, overflow: isFullscreen ? 'hidden' : undefined, paddingX: 0, children: [grouped.length === 0 ? (_jsx(Text, { color: tokens.colors.dim, children: "Message Klyro..." })) : visibleGrouped.map((item) => {
                                 if (item.verb) {
                                     const gr = item;
                                     const isExpanded = expandedGroups.has(gr.id);
@@ -276,8 +365,8 @@ export function App(props) {
                                             return `Edited ${gr.items.length} files`;
                                         return `${gr.verb} ${gr.items.length} items`;
                                     })();
-                                    const right = gr.status === 'running' ? `${(elapsed / 1000).toFixed(1)}s` : gr.status === 'error' ? '✗' : `${gr.totalMs}ms`;
-                                    const marker = isExpanded ? '▼' : '✓';
+                                    const right = gr.status === 'running' ? `${(elapsed / 1000).toFixed(1)}s` : gr.status === 'error' ? 'œ—' : `${gr.totalMs}ms`;
+                                    const marker = isExpanded ? '–¼' : 'œ“';
                                     const markerColor = gr.status === 'error' ? tokens.colors.err : gr.status === 'running' ? tokens.colors.warn : tokens.colors.ok;
                                     return (_jsxs(Box, { flexDirection: "column", marginBottom: 1, children: [_jsxs(Box, { children: [_jsxs(Text, { color: tokens.colors.guide, children: ["  ", g('guide'), "   "] }), _jsxs(Text, { color: markerColor, children: [marker, " ", verbLine] }), _jsxs(Text, { color: tokens.colors.dim, children: ["  ", right] })] }), isExpanded ? gr.items.map((it) => {
                                                 let friendly = '';
@@ -305,11 +394,11 @@ export function App(props) {
                                     return _jsxs(Box, { marginBottom: 1, children: [_jsxs(Text, { color: tokens.colors.accent, bold: true, children: [g('prompt'), " "] }), _jsx(Text, { wrap: "wrap", children: it.text })] }, it.id);
                                 }
                                 if (it.kind === 'text') {
-                                    // prose — render markdown, not raw **, with proper wrap and guide
+                                    // prose — render markdown, not raw **, with proper wrap and guide
                                     return (_jsxs(Box, { flexDirection: "column", marginBottom: 1, children: [_jsxs(Box, { children: [_jsxs(Text, { color: tokens.colors.guide, children: ["  ", g('guide'), "   "] }), _jsxs(Text, { color: tokens.colors.accent, children: [g('agentBullet'), " Klyro"] })] }), _jsx(Box, { paddingLeft: 2, flexDirection: "column", children: _jsxs(Box, { children: [_jsxs(Text, { color: tokens.colors.dim, children: ["  ", g('guide'), "   "] }), _jsx(Box, { flexGrow: 1, children: _jsx(MarkdownText, { text: it.text }) })] }) })] }, it.id));
                                 }
                                 if (it.kind === 'error')
-                                    return _jsx(Box, { paddingLeft: 2, marginBottom: 1, children: _jsxs(Text, { color: tokens.colors.err, children: ["  ", g('guide'), "   \u00E2\u0153\u2014 ", it.message] }) }, it.id);
+                                    return _jsx(Box, { paddingLeft: 2, marginBottom: 1, children: _jsxs(Text, { color: tokens.colors.err, children: ["  ", g('guide'), "   \u0153\u2014 ", it.message] }) }, it.id);
                                 if (it.kind === 'policy')
                                     return null;
                                 if (it.kind === 'file_changed')
@@ -317,5 +406,5 @@ export function App(props) {
                                 if (it.kind === 'diff')
                                     return (_jsxs(Box, { flexDirection: "column", paddingLeft: 2, marginBottom: 1, children: [_jsx(Text, { bold: true, color: tokens.colors.soft, children: it.summary ?? 'Diff' }), it.hunks.map((h, i) => (_jsxs(Box, { flexDirection: "column", marginTop: 0, children: [_jsx(Text, { color: tokens.colors.soft, children: h.path }), h.lines.map((l, j) => (_jsxs(Text, { wrap: "wrap", color: l.kind === 'add' ? tokens.colors.ok : l.kind === 'remove' ? tokens.colors.err : tokens.colors.dim, children: [l.kind === 'add' ? '+ ' : l.kind === 'remove' ? '- ' : '  ', l.text] }, j)))] }, i)))] }, it.id));
                                 return null;
-                            }), status.status === 'running' && !streamingIdRef.current ? (_jsxs(Box, { paddingLeft: 2, marginBottom: 1, children: [_jsxs(Text, { color: tokens.colors.guide, children: ["  ", g('guide'), "   "] }), _jsx(Text, { color: tokens.colors.dim, children: "Thinking..." }), _jsxs(Text, { color: tokens.colors.dim, children: ["  ", (elapsed / 1000).toFixed(1), "s"] })] })) : null, plan.length > 0 ? (_jsxs(Box, { flexDirection: "column", paddingLeft: 2, marginTop: 0, marginBottom: 1, children: [_jsxs(Box, { children: [_jsxs(Text, { color: tokens.colors.guide, children: ["  ", g('guide'), "   "] }), _jsxs(Text, { bold: true, children: [g('todoPlan'), " Plan  ", plan.filter((p) => p.status === 'done').length, "/", plan.length] })] }), plan.slice(0, 8).map((p) => (_jsxs(Box, { children: [_jsxs(Text, { color: tokens.colors.guide, children: ["  ", g('guide'), "   "] }), _jsxs(Text, { color: p.status === 'done' ? tokens.colors.ok : p.status === 'in_progress' ? tokens.colors.accent : tokens.colors.dim, children: [p.status === 'done' ? g('todoDone') : p.status === 'in_progress' ? g('todoActive') : g('todoPending'), " ", p.title] })] }, p.id)))] })) : null, queuedInputs.length > 0 ? (_jsx(Box, { flexDirection: "column", paddingLeft: 2, marginBottom: 1, children: queuedInputs.map((q, i) => (_jsxs(Text, { color: tokens.colors.dim, children: ["queued: ", q.slice(0, 60), i === 0 ? '  esc to drop' : ''] }, i))) })) : null] }), isFullscreen ? (_jsx(Box, { flexDirection: "column", width: 1, marginLeft: 1, children: Array.from({ length: trackH }).map((_, i) => (_jsx(Text, { color: i === thumbPos ? tokens.colors.accent : tokens.colors.guide, children: i === thumbPos ? '●' : '│' }, i))) })) : null] }), _jsxs(Box, { flexDirection: "column", children: [_jsx(Text, { color: tokens.colors.guide, children: g('rule').repeat(Math.max(10, width - 2)) }), _jsxs(Box, { children: [_jsxs(Text, { color: tokens.colors.accent, bold: true, children: [g('prompt'), " "] }), _jsxs(Text, { wrap: "wrap", children: [input || _jsx(Text, { color: tokens.colors.dim, children: "Message Klyro\u00E2\u20AC\u00A6" }), "\u00E2\u2013\u008F"] })] }), _jsx(Text, { color: tokens.colors.guide, children: g('rule').repeat(Math.max(10, width - 2)) })] }), _jsxs(Box, { justifyContent: "space-between", children: [_jsxs(Text, { color: tokens.colors.dim, children: [baseHints, maxOffset > 0 && isFullscreen ? '  ·  PgUp/Dn scroll' : ''] }), _jsxs(Text, { color: tokens.colors.dim, children: [cost > 0 ? `$${cost.toFixed(2)} · ` : '', ctxPct > 0 ? `${ctxPct}% ctx · ` : '', status.status === 'running' ? 'auto mode on ●' : ''] })] })] }));
+                            }), status.status === 'running' && !streamingIdRef.current ? (_jsxs(Box, { paddingLeft: 2, marginBottom: 1, children: [_jsxs(Text, { color: tokens.colors.guide, children: ["  ", g('guide'), "   "] }), _jsx(Text, { color: tokens.colors.dim, children: "Thinking..." }), _jsxs(Text, { color: tokens.colors.dim, children: ["  ", (elapsed / 1000).toFixed(1), "s"] })] })) : null, plan.length > 0 ? (_jsxs(Box, { flexDirection: "column", paddingLeft: 2, marginTop: 0, marginBottom: 1, children: [_jsxs(Box, { children: [_jsxs(Text, { color: tokens.colors.guide, children: ["  ", g('guide'), "   "] }), _jsxs(Text, { bold: true, children: [g('todoPlan'), " Plan  ", plan.filter((p) => p.status === 'done').length, "/", plan.length] })] }), plan.slice(0, 8).map((p) => (_jsxs(Box, { children: [_jsxs(Text, { color: tokens.colors.guide, children: ["  ", g('guide'), "   "] }), _jsxs(Text, { color: p.status === 'done' ? tokens.colors.ok : p.status === 'in_progress' ? tokens.colors.accent : tokens.colors.dim, children: [p.status === 'done' ? g('todoDone') : p.status === 'in_progress' ? g('todoActive') : g('todoPending'), " ", p.title] })] }, p.id)))] })) : null, queuedInputs.length > 0 ? (_jsx(Box, { flexDirection: "column", paddingLeft: 2, marginBottom: 1, children: queuedInputs.map((q, i) => (_jsxs(Text, { color: tokens.colors.dim, children: ["queued: ", q.slice(0, 60), i === 0 ? '  esc to drop' : ''] }, i))) })) : null] }), isFullscreen ? (_jsx(Box, { flexDirection: "column", width: 1, marginLeft: 1, children: Array.from({ length: trackH }).map((_, i) => (_jsx(Text, { color: i === thumbPos ? tokens.colors.accent : tokens.colors.guide, children: i === thumbPos ? '●' : '│' }, i))) })) : null] }), pinned && pendingNew > 0 ? (_jsx(Box, { justifyContent: "flex-end", paddingX: 1, marginTop: -1, children: _jsxs(Text, { backgroundColor: tokens.colors.accentSoft, color: tokens.colors.accent, bold: true, children: [' ↓ ', pendingNew, ' new ', pendingNew === 1 ? 'message' : 'messages', ' '] }) })) : null, _jsxs(Box, { flexDirection: "column", children: [_jsx(Text, { color: tokens.colors.guide, children: g('rule').repeat(Math.max(10, width - 2)) }), _jsxs(Box, { children: [_jsxs(Text, { color: tokens.colors.accent, bold: true, children: [g('prompt'), " "] }), _jsxs(Text, { wrap: "wrap", children: [input || _jsx(Text, { color: tokens.colors.dim, children: "Message Klyro..." }), "|"] })] }), _jsx(Text, { color: tokens.colors.guide, children: g('rule').repeat(Math.max(10, width - 2)) })] }), _jsxs(Box, { justifyContent: "space-between", children: [_jsxs(Text, { color: tokens.colors.dim, children: [baseHints, maxOffset > 0 && isFullscreen ? '  ·  PgUp/Dn scroll' : ''] }), _jsxs(Text, { color: tokens.colors.dim, children: [cost > 0 ? `$${cost.toFixed(2)} · ` : '', ctxPct > 0 ? `${ctxPct}% ctx · ` : '', status.status === 'running' ? 'auto mode on ●' : ''] })] })] }));
 }

package/dist/tui/app.test.js

@@ -79,4 +79,146 @@ describe('App', () => {
         const call = onSlash.mock.calls[0]?.[0];
         expect(call?.kind).toBe('quit');
     });
+    // --- Chat scroll behavior (TUI_DESIGN chat_scroll.md) -----------------
+    // Build a 25-item initial transcript. Each item has a unique tag so we can
+    // grep `lastFrame()` for it.
+    function makeInitialTranscript(n) {
+        const out = [];
+        for (let i = 0; i < n; i++) {
+            out.push({
+                id: `seed-${i}`,
+                kind: 'text',
+                text: `MSG-${i.toString().padStart(2, '0')}-tag`,
+                role: 'user',
+            });
+        }
+        return out;
+    }
+    // ANSI sequences Ink's parse-keypress recognizes.
+    const KEY_HOME = '\x1b[H';
+    const KEY_END = '\x1b[F';
+    const KEY_PGUP = '\x1b[5~';
+    const KEY_PGDN = '\x1b[6~';
+    const KEY_SHIFT_UP = '\x1b[1;2A';
+    const KEY_SHIFT_DOWN = '\x1b[1;2B';
+    it('starts at the bottom (follow-tail) when initial content fills the viewport', async () => {
+        const { lastFrame } = render(_jsx(App, { initialModel: "m", maxSteps: 10, cwd: "/test", onPrompt: async () => { }, onSlash: async () => { }, isFullscreen: true, initialTranscript: makeInitialTranscript(25) }));
+        await new Promise((r) => setTimeout(r, 50));
+        const frame = lastFrame() ?? '';
+        // The viewport is 20 rows; the last few seeded items (MSG-22..MSG-24) should
+        // be in the visible window. The first item (MSG-00) should NOT be visible.
+        expect(frame).toMatch(/MSG-24-tag/);
+        expect(frame).toMatch(/MSG-23-tag/);
+        expect(frame).not.toMatch(/MSG-00-tag/);
+    });
+    it('Home jumps to the top; End re-engages follow-tail', async () => {
+        const { stdin, lastFrame } = render(_jsx(App, { initialModel: "m", maxSteps: 10, cwd: "/test", onPrompt: async () => { }, onSlash: async () => { }, isFullscreen: true, initialTranscript: makeInitialTranscript(25) }));
+        await new Promise((r) => setTimeout(r, 50));
+        stdin.write(KEY_HOME);
+        await new Promise((r) => setTimeout(r, 30));
+        const top = lastFrame() ?? '';
+        expect(top).toMatch(/MSG-00-tag/);
+        expect(top).not.toMatch(/MSG-24-tag/);
+        // End re-engages follow-tail.
+        stdin.write(KEY_END);
+        await new Promise((r) => setTimeout(r, 30));
+        const bottom = lastFrame() ?? '';
+        expect(bottom).toMatch(/MSG-24-tag/);
+        expect(bottom).not.toMatch(/MSG-00-tag/);
+    });
+    it('PageUp/PageDown snap to message boundaries', async () => {
+        const { stdin, lastFrame } = render(_jsx(App, { initialModel: "m", maxSteps: 10, cwd: "/test", onPrompt: async () => { }, onSlash: async () => { }, isFullscreen: true, initialTranscript: makeInitialTranscript(25) }));
+        await new Promise((r) => setTimeout(r, 50));
+        // Go to top, then PageDown 3 times. Each PageDown should land on a message
+        // boundary, so visible window starts at one of the seeded indices.
+        stdin.write(KEY_HOME);
+        await new Promise((r) => setTimeout(r, 30));
+        stdin.write(KEY_PGDN);
+        await new Promise((r) => setTimeout(r, 30));
+        stdin.write(KEY_PGDN);
+        await new Promise((r) => setTimeout(r, 30));
+        stdin.write(KEY_PGDN);
+        await new Promise((r) => setTimeout(r, 30));
+        const frame = lastFrame() ?? '';
+        // After 3 PageDowns from top, the earliest visible item should be MSG-03
+        // (snap-to-message keeps the boundary on the first visible row). We assert
+        // that MSG-03 is visible and MSG-00 is not.
+        expect(frame).toMatch(/MSG-03-tag/);
+        expect(frame).not.toMatch(/MSG-00-tag/);
+    });
+    it('pins to top: new content does NOT auto-scroll when user has scrolled up', async () => {
+        const { stdin, lastFrame } = render(_jsx(App, { initialModel: "m", maxSteps: 10, cwd: "/test", onPrompt: async () => { }, onSlash: async () => { }, isFullscreen: true, initialTranscript: makeInitialTranscript(25) }));
+        await new Promise((r) => setTimeout(r, 100));
+        // Pin: scroll up to top.
+        stdin.write(KEY_HOME);
+        await new Promise((r) => setTimeout(r, 100));
+        const before = lastFrame() ?? '';
+        expect(before).toMatch(/MSG-00-tag/);
+        expect(before).not.toMatch(/MSG-24-tag/);
+        // New content arrives while pinned.
+        const g = globalThis;
+        g.__klyroAppAppend({
+            id: 'late-1',
+            kind: 'text',
+            text: 'LATE-1-tag',
+            role: 'assistant',
+        });
+        g.__klyroAppAppend({
+            id: 'late-2',
+            kind: 'text',
+            text: 'LATE-2-tag',
+            role: 'assistant',
+        });
+        g.__klyroAppAppend({
+            id: 'late-3',
+            kind: 'text',
+            text: 'LATE-3-tag',
+            role: 'assistant',
+        });
+        await new Promise((r) => setTimeout(r, 200));
+        const after = lastFrame() ?? '';
+        // Still pinned at top: MSG-00 visible, LATE items not in viewport.
+        expect(after).toMatch(/MSG-00-tag/);
+        expect(after).not.toMatch(/LATE-1-tag/);
+    });
+    it('pressing End re-engages follow-tail and reveals new content', async () => {
+        const { stdin, lastFrame } = render(_jsx(App, { initialModel: "m", maxSteps: 10, cwd: "/test", onPrompt: async () => { }, onSlash: async () => { }, isFullscreen: true, initialTranscript: makeInitialTranscript(25) }));
+        await new Promise((r) => setTimeout(r, 100));
+        stdin.write(KEY_HOME);
+        await new Promise((r) => setTimeout(r, 100));
+        const g = globalThis;
+        g.__klyroAppAppend({
+            id: 'late-1',
+            kind: 'text',
+            text: 'LATE-1-tag',
+            role: 'assistant',
+        });
+        await new Promise((r) => setTimeout(r, 200));
+        expect(lastFrame() ?? '').not.toMatch(/LATE-1-tag/);
+        // End re-engages follow-tail and shows the new content.
+        stdin.write(KEY_END);
+        await new Promise((r) => setTimeout(r, 100));
+        const frame = lastFrame() ?? '';
+        expect(frame).toMatch(/LATE-1-tag/);
+    });
+    it('Shift+Up / Shift+Down scroll by one line', async () => {
+        const { stdin, lastFrame } = render(_jsx(App, { initialModel: "m", maxSteps: 10, cwd: "/test", onPrompt: async () => { }, onSlash: async () => { }, isFullscreen: true, initialTranscript: makeInitialTranscript(25) }));
+        await new Promise((r) => setTimeout(r, 50));
+        // Jump to top, then shift+down a few times, then back with shift+up.
+        stdin.write(KEY_HOME);
+        await new Promise((r) => setTimeout(r, 30));
+        stdin.write(KEY_SHIFT_DOWN);
+        stdin.write(KEY_SHIFT_DOWN);
+        await new Promise((r) => setTimeout(r, 30));
+        const frame = lastFrame() ?? '';
+        // Shift+Down from scrollOffset=0 moves us down by 2. The visible window
+        // is now [2..22). MSG-00 should be off-screen, MSG-02 should be on-screen.
+        expect(frame).not.toMatch(/MSG-00-tag/);
+        expect(frame).toMatch(/MSG-02-tag/);
+        // Shift+Up once: scrollOffset back to 1, MSG-01 visible, MSG-02 still visible.
+        stdin.write(KEY_SHIFT_UP);
+        await new Promise((r) => setTimeout(r, 30));
+        const frame2 = lastFrame() ?? '';
+        expect(frame2).toMatch(/MSG-01-tag/);
+    });
 });

package/dist/verification/baseline.js

@@ -66,17 +66,11 @@ export async function runBaseline(cwd, command, timeoutMs = 90_000) {
     return baseline;
 }
 const MAX_BASELINE_BYTES = 256 * 1024;
-function cap(cur, chunk) {
-    if (cur.length >= MAX_BASELINE_BYTES)
-        return cur;
-    const n = cur + chunk;
-    return n.length > MAX_BASELINE_BYTES ? n.slice(0, MAX_BASELINE_BYTES) + '\n... [truncated]' : n;
-}
 function runCmd(cwd, command, timeoutMs) {
     return new Promise((resolve) => {
         const child = spawn(command, { cwd, shell: true, env: process.env });
-        let stdout = '';
-        let stderr = '';
+        const outChunks = [];
+        const errChunks = [];
         let done = false;
         const timer = setTimeout(() => {
             if (done)
@@ -86,24 +80,31 @@ function runCmd(cwd, command, timeoutMs) {
                 child.kill();
             }
             catch { /* ignore */ }
-            resolve({ ok: false, exitCode: -1, stdout, stderr: stderr + '\n[baseline timeout]' });
+            const so = Buffer.concat(outChunks).toString('utf-8').slice(0, MAX_BASELINE_BYTES);
+            const se = Buffer.concat(errChunks).toString('utf-8').slice(0, MAX_BASELINE_BYTES);
+            resolve({ ok: false, exitCode: -1, stdout: so, stderr: se + '\n[baseline timeout]' });
         }, timeoutMs);
-        child.stdout.on('data', (b) => { stdout = cap(stdout, b.toString()); });
-        child.stderr.on('data', (b) => { stderr = cap(stderr, b.toString()); });
+        child.stdout.on('data', (b) => { if (Buffer.concat(outChunks).length < MAX_BASELINE_BYTES)
+            outChunks.push(b); });
+        child.stderr.on('data', (b) => { if (Buffer.concat(errChunks).length < MAX_BASELINE_BYTES)
+            errChunks.push(b); });
         child.on('close', (code) => {
             if (done)
                 return;
             done = true;
             clearTimeout(timer);
             const exit = typeof code === 'number' ? code : -1;
-            resolve({ ok: exit === 0, exitCode: exit, stdout, stderr });
+            const so = Buffer.concat(outChunks).toString('utf-8').slice(0, MAX_BASELINE_BYTES);
+            const se = Buffer.concat(errChunks).toString('utf-8').slice(0, MAX_BASELINE_BYTES);
+            resolve({ ok: exit === 0, exitCode: exit, stdout: so, stderr: se });
         });
         child.on('error', (err) => {
             if (done)
                 return;
             done = true;
             clearTimeout(timer);
-            resolve({ ok: false, exitCode: -1, stdout, stderr: String(err) });
+            const so = Buffer.concat(outChunks).toString('utf-8').slice(0, MAX_BASELINE_BYTES);
+            resolve({ ok: false, exitCode: -1, stdout: so, stderr: String(err) });
         });
     });
 }

package/dist/verification/classify.js

@@ -7,7 +7,7 @@ import { spawn } from 'node:child_process';
 export function classifyFailure(current, baseline, flakyRerunOk) {
     const combined = (current.stderr + '\n' + current.stdout).toLowerCase();
     // env: missing binary, network, permission, no such file, EACCES etc
-    if (/enoent|command not found|no such file|network|econn|etimedout|eacces|permission denied|env/.test(combined) && /error/i.test(combined)) {
+    if (/enoent|command not found|no such file|network|econn|etimedout|eacces|permission denied|environment variable/.test(combined) && /error/i.test(combined)) {
         // only if not a real test failure but env
         if (!current.failure || current.failure.type === 'unknown')
             return 'env';
@@ -42,7 +42,7 @@ export async function rerunOnce(cwd, command, timeoutMs = 45_000) {
         const t = setTimeout(() => { if (!done) {
             done = true;
             try {
-                child.kill();
+                child.kill('SIGKILL');
             }
             catch { }
             resolve(false);
@@ -72,19 +72,30 @@ export async function gatherRepairContext(cwd, failure) {
         }
     }
     // hunks: git diff --stat + --unified=2 for changed files
-    const hunks = await execCapture(cwd, 'git diff --stat && echo "---" && git diff -U2 2>&1 | head -n 300');
-    const blame = failure?.files[0]?.path
-        ? await execCapture(cwd, `git blame "${failure.files[0].path}" 2>&1 | head -n 20`)
+    const hunks = await execCapturePipe(cwd, 'git diff --stat && echo "---" && git diff -U2 2>&1 | head -n 300');
+    const blamePath = failure?.files[0]?.path;
+    const blame = blamePath && /^[\w./-]+$/.test(blamePath) && !blamePath.includes('..')
+        ? await execCaptureArgv(cwd, blamePath, ['--no-pager', 'blame', '--'])
         : '';
     return { failingTests, hunks, blame };
 }
-function execCapture(cwd, cmd) {
+function execCapturePipe(cwd, cmd) {
     return new Promise((resolve) => {
         const child = spawn(cmd, { cwd, shell: true, env: process.env });
-        let out = '';
-        child.stdout.on('data', (b) => { out += b.toString(); });
-        child.stderr.on('data', (b) => { out += b.toString(); });
-        child.on('close', () => resolve(out.slice(0, 4000)));
+        const chunks = [];
+        child.stdout.on('data', (b) => { chunks.push(b); });
+        child.stderr.on('data', (b) => { chunks.push(b); });
+        child.on('close', () => resolve(Buffer.concat(chunks).toString().slice(0, 4000)));
+        child.on('error', () => resolve(''));
+    });
+}
+function execCaptureArgv(cwd, file, args) {
+    return new Promise((resolve) => {
+        const child = spawn('git', [...args, file], { cwd, shell: false, env: process.env });
+        const chunks = [];
+        child.stdout.on('data', (b) => { chunks.push(b); });
+        child.stderr.on('data', (b) => { chunks.push(b); });
+        child.on('close', () => resolve(Buffer.concat(chunks).toString().split('\n').slice(0, 20).join('\n')));
         child.on('error', () => resolve(''));
     });
 }

package/dist/verification/engine.js

@@ -17,41 +17,88 @@ function appendCapped(current, chunk) {
     const next = current + chunk;
     return next.length > MAX_VERIFY_BYTES ? next.slice(0, MAX_VERIFY_BYTES) + '\n... [truncated]' : next;
 }
+// SEC-004: denylist for dangerous patterns in verify commands.
+// Reuses the same patterns as shell-exec.ts DANGEROUS_PATTERNS.
+const DANGEROUS_VERIFY_PATTERNS = [
+    { pattern: /rm\s+-rf?\s+\//, reason: 'recursive delete at filesystem root' },
+    { pattern: /rm\s+-rf?\s+\/\/+/, reason: 'recursive delete at filesystem root (//)' },
+    { pattern: /rm\s+-rf?\s+\/\*/, reason: 'recursive delete at filesystem root (/*)' },
+    { pattern: /rm\s+-rf?\s+\.\s*($|[;&|])/, reason: 'recursive delete current directory' },
+    { pattern: /rm\s+-rf?\s+\*\s*($|[;&|])/, reason: 'recursive delete all files via *' },
+    { pattern: /rm\s+-rf?\s+\.\/\*\s*($|[;&|])/, reason: 'recursive delete all files' },
+    { pattern: /rm\s+-rf?\s+~(\/|$)/, reason: 'recursive delete home directory via ~' },
+    { pattern: /rm\s+-rf?\s+\$HOME\b/, reason: 'recursive delete home via $HOME' },
+    { pattern: /rm\s+-rf?\s+\$PWD\b/, reason: 'recursive delete via $PWD' },
+    { pattern: /del\s+\/s\s+\/q\s+[a-z]:\\/i, reason: 'recursive delete on Windows drive root' },
+    { pattern: /:\(\)\s*\{\s*:\s*\|\s*:\s*&\s*\}\s*;\s*:/, reason: 'fork bomb' },
+    { pattern: /bomb\(\)\s*\{\s*bomb\|bomb/, reason: 'fork bomb variant' },
+    { pattern: />\s*\/dev\/sd[a-z]/, reason: 'overwrite raw block device' },
+    { pattern: /mkfs(\.|\s)/, reason: 'format filesystem' },
+    { pattern: /dd\s+.*of=\/dev\//, reason: 'dd write to device' },
+    { pattern: /chmod\s+-R\s+777\s+\//, reason: 'chmod 777 on root' },
+    { pattern: /curl.*\|\s*(sh|bash|zsh|python|python3|perl|ruby|php)/i, reason: 'curl|sh to unknown host' },
+    { pattern: /wget.*\|\s*(sh|bash|python|perl|ruby)/i, reason: 'wget|sh pipe' },
+    { pattern: /rm\s+-rf\s+--no-preserve-root\s+\//, reason: 'recursive delete --no-preserve-root' },
+    { pattern: /\$\(/, reason: 'command substitution $()' },
+    { pattern: /`[^`]*`/, reason: 'command substitution via backticks' },
+    { pattern: /\|\s*bash\b|\|\s*sh\b/, reason: 'pipe to shell' },
+    { pattern: /;\s*rm\s+-rf/, reason: 'chained rm -rf' },
+    { pattern: /&&\s*rm\s+-rf/, reason: 'chained rm -rf' },
+    { pattern: /\|\|\s*rm\s+-rf/, reason: 'chained rm -rf' },
+];
 export async function verify(opts) {
+    // SEC-004: reject dangerous patterns before spawning
+    for (const { pattern, reason } of DANGEROUS_VERIFY_PATTERNS) {
+        if (pattern.test(opts.command)) {
+            return {
+                ok: false,
+                exitCode: -1,
+                stdout: '',
+                stderr: `Command blocked: ${reason}`,
+                failure: { type: 'runtime', files: [], raw: `Command blocked: ${reason}`, exitCode: -1 },
+            };
+        }
+    }
     const timeout = opts.timeoutMs ?? 5 * 60 * 1000;
     return new Promise((resolve) => {
         const child = spawn(opts.command, { cwd: opts.cwd, shell: true, env: process.env });
-        let stdout = '';
-        let stderr = '';
+        const outChunks = [];
+        const errChunks = [];
         let done = false;
         const timer = setTimeout(() => {
             if (done)
                 return;
             child.kill();
             done = true;
-            const raw = stderr + '\n' + stdout;
+            const so = Buffer.concat(outChunks).toString('utf-8').slice(0, MAX_VERIFY_BYTES);
+            const se = Buffer.concat(errChunks).toString('utf-8').slice(0, MAX_VERIFY_BYTES);
+            const raw = se + '\n' + so;
             resolve({
                 ok: false,
                 exitCode: -1,
-                stdout,
-                stderr: stderr + '\n[verify timeout]',
+                stdout: so,
+                stderr: se + '\n[verify timeout]',
                 failure: { type: 'runtime', files: [], raw, exitCode: -1 },
             });
         }, timeout);
-        child.stdout.on('data', (b) => { stdout = appendCapped(stdout, b.toString()); });
-        child.stderr.on('data', (b) => { stderr = appendCapped(stderr, b.toString()); });
+        child.stdout.on('data', (b) => { if (Buffer.concat(outChunks).length < MAX_VERIFY_BYTES)
+            outChunks.push(b); });
+        child.stderr.on('data', (b) => { if (Buffer.concat(errChunks).length < MAX_VERIFY_BYTES)
+            errChunks.push(b); });
         child.on('close', (code) => {
             if (done)
                 return;
             done = true;
             clearTimeout(timer);
+            const so = Buffer.concat(outChunks).toString('utf-8').slice(0, MAX_VERIFY_BYTES);
+            const se = Buffer.concat(errChunks).toString('utf-8').slice(0, MAX_VERIFY_BYTES);
             const exit = typeof code === 'number' ? code : -1;
             if (exit === 0) {
-                resolve({ ok: true, exitCode: 0, stdout, stderr });
+                resolve({ ok: true, exitCode: 0, stdout: so, stderr: se });
                 return;
             }
-            const failure = detect(stdout, stderr, exit);
-            resolve({ ok: false, exitCode: exit, stdout, stderr, failure });
+            const failure = detect(so, se, exit);
+            resolve({ ok: false, exitCode: exit, stdout: so, stderr: se, failure });
         });
     });
 }

package/dist/verification/scoped.js

@@ -71,17 +71,11 @@ export function buildScopedCommand(cwd, baseCommand, relatedTests) {
     return null;
 }
 const MAX_SCOPED_BYTES = 256 * 1024;
-function appendCappedScoped(cur, chunk) {
-    if (cur.length >= MAX_SCOPED_BYTES)
-        return cur;
-    const n = cur + chunk;
-    return n.length > MAX_SCOPED_BYTES ? n.slice(0, MAX_SCOPED_BYTES) + '\n... [truncated]' : n;
-}
 export async function runScopedVerify(cwd, command, timeoutMs = 45_000) {
     return new Promise((resolve) => {
         const child = spawn(command, { cwd, shell: true, env: process.env });
-        let stdout = '';
-        let stderr = '';
+        const outChunks = [];
+        const errChunks = [];
         let done = false;
         const timer = setTimeout(() => {
             if (done)
@@ -91,24 +85,31 @@ export async function runScopedVerify(cwd, command, timeoutMs = 45_000) {
                 child.kill();
             }
             catch { /* ignore */ }
-            resolve({ ok: false, exitCode: -1, stdout, stderr: stderr + '\n[scoped timeout]' });
+            const so = Buffer.concat(outChunks).toString('utf-8').slice(0, MAX_SCOPED_BYTES);
+            const se = Buffer.concat(errChunks).toString('utf-8').slice(0, MAX_SCOPED_BYTES);
+            resolve({ ok: false, exitCode: -1, stdout: so, stderr: se + '\n[scoped timeout]' });
         }, timeoutMs);
-        child.stdout.on('data', (b) => { stdout = appendCappedScoped(stdout, b.toString()); });
-        child.stderr.on('data', (b) => { stderr = appendCappedScoped(stderr, b.toString()); });
+        child.stdout.on('data', (b) => { if (Buffer.concat(outChunks).length < MAX_SCOPED_BYTES)
+            outChunks.push(b); });
+        child.stderr.on('data', (b) => { if (Buffer.concat(errChunks).length < MAX_SCOPED_BYTES)
+            errChunks.push(b); });
         child.on('close', (code) => {
             if (done)
                 return;
             done = true;
             clearTimeout(timer);
             const exit = typeof code === 'number' ? code : -1;
-            resolve({ ok: exit === 0, exitCode: exit, stdout, stderr });
+            const so = Buffer.concat(outChunks).toString('utf-8').slice(0, MAX_SCOPED_BYTES);
+            const se = Buffer.concat(errChunks).toString('utf-8').slice(0, MAX_SCOPED_BYTES);
+            resolve({ ok: exit === 0, exitCode: exit, stdout: so, stderr: se });
         });
         child.on('error', (err) => {
             if (done)
                 return;
             done = true;
             clearTimeout(timer);
-            resolve({ ok: false, exitCode: -1, stdout, stderr: String(err) });
+            const so = Buffer.concat(outChunks).toString('utf-8').slice(0, MAX_SCOPED_BYTES);
+            resolve({ ok: false, exitCode: -1, stdout: so, stderr: String(err) });
         });
     });
 }

package/package.json

@@ -1,60 +1,60 @@
-{
-  "name": "klyro",
-  "version": "0.1.40",
-  "description": "Klyro — autonomous coding harness CLI that streams from any OpenAI-compatible or Anthropic LLM endpoint.",
-  "type": "module",
-  "main": "dist/index.js",
-  "types": "dist/index.d.ts",
-  "bin": {
-    "klyro": "dist/index.js",
-    "ky": "dist/index.js"
-  },
-  "files": [
-    "dist",
-    "README.md",
-    "READ.md",
-    "LICENSE"
-  ],
-  "engines": {
-    "node": ">=20"
-  },
-  "scripts": {
-    "build": "tsc",
-    "start": "node dist/index.js",
-    "dev": "tsx src/index.ts",
-    "dev:watch": "tsx watch src/index.ts",
-    "typecheck": "tsc --noEmit",
-    "test": "vitest run",
-    "test:watch": "vitest",
-    "prepublishOnly": "npm run typecheck && npm test && npm run build",
-    "pack:dry": "npm pack --dry-run",
-    "publish:public": "npm publish --access public"
-  },
-  "keywords": [
-    "llm",
-    "ai",
-    "cli",
-    "openai",
-    "anthropic",
-    "harness",
-    "agent",
-    "streaming",
-    "klyro"
-  ],
-  "license": "MIT",
-  "dependencies": {
-    "commander": "^12.1.0",
-    "ink": "^7.1.1",
-    "ink-spinner": "^5.0.0",
-    "react": "^19.2.0",
-    "zod": "^4.5.4"
-  },
-  "devDependencies": {
-    "@types/node": "^22.9.0",
-    "@types/react": "^19.2.18",
-    "ink-testing-library": "^4.0.0",
-    "tsx": "^4.23.13",
-    "typescript": "^5.5.4",
-    "vitest": "^4.1.11"
-  }
-}
+{
+  "name": "klyro",
+  "version": "0.1.42",
+  "description": "Klyro \u2014 autonomous coding harness CLI that streams from any OpenAI-compatible or Anthropic LLM endpoint.",
+  "type": "module",
+  "main": "dist/index.js",
+  "types": "dist/index.d.ts",
+  "bin": {
+    "klyro": "dist/index.js",
+    "ky": "dist/index.js"
+  },
+  "files": [
+    "dist",
+    "README.md",
+    "READ.md",
+    "LICENSE"
+  ],
+  "engines": {
+    "node": ">=20"
+  },
+  "scripts": {
+    "build": "tsc",
+    "start": "node dist/index.js",
+    "dev": "tsx src/index.ts",
+    "dev:watch": "tsx watch src/index.ts",
+    "typecheck": "tsc --noEmit",
+    "test": "vitest run",
+    "test:watch": "vitest",
+    "prepublishOnly": "npm run typecheck && npm test && npm run build",
+    "pack:dry": "npm pack --dry-run",
+    "publish:public": "npm publish --access public"
+  },
+  "keywords": [
+    "llm",
+    "ai",
+    "cli",
+    "openai",
+    "anthropic",
+    "harness",
+    "agent",
+    "streaming",
+    "klyro"
+  ],
+  "license": "MIT",
+  "dependencies": {
+    "commander": "^12.1.0",
+    "ink": "^7.1.1",
+    "ink-spinner": "^5.0.0",
+    "react": "^19.2.0",
+    "zod": "^4.5.4"
+  },
+  "devDependencies": {
+    "@types/node": "^22.9.0",
+    "@types/react": "^19.2.18",
+    "ink-testing-library": "^4.0.0",
+    "tsx": "^4.23.13",
+    "typescript": "^5.5.4",
+    "vitest": "^4.1.11"
+  }
+}