pi-harness-delegate 0.2.1 → 0.3.0

package/extensions/harnesses/amp.ts

@@ -1,4 +1,6 @@
 import { execFile } from 'node:child_process';
+import { accessSync, constants } from 'node:fs';
+import { delimiter, join } from 'node:path';
 import { promisify } from 'node:util';
 import type {
   BuildArgsOpts,
@@ -9,87 +11,205 @@ import type {
   StreamedResult,
 } from './types.ts';
 
-const execFileAsync = promisify(execFile);
+// Schema verified against omp/17.2.9 (the only binary of this harness installed on the capture
+// machine — see tests/fixtures/amp.jsonl and AGENTS.md for what "amp" vs "omp" means here).
 
+const execFileAsync = promisify(execFile);
 function isRecord(v: unknown): v is Record<string, unknown> {
   return typeof v === 'object' && v !== null && !Array.isArray(v);
 }
 
+/**
+ * `amp` isn't on PATH on machines that only have the `omp` alias binary installed — resolve
+ * which one actually exists once at module load, so the spawned binary matches what `detect()`
+ * already tolerates. Falls back to 'amp' (the documented default) when neither is found, so
+ * error messages still name the expected tool.
+ */
+export function resolveAmpBinary(pathEnv: string | undefined, exists: (p: string) => boolean = pathExists): string {
+  const dirs = (pathEnv ?? '').split(delimiter).filter(Boolean);
+  for (const dir of dirs) {
+    if (exists(join(dir, 'amp'))) return 'amp';
+  }
+  for (const dir of dirs) {
+    if (exists(join(dir, 'omp'))) return 'omp';
+  }
+  return 'amp';
+}
+
+function pathExists(p: string): boolean {
+  try {
+    accessSync(p, constants.X_OK);
+    return true;
+  } catch {
+    return false;
+  }
+}
+
+const RESOLVED_BINARY = resolveAmpBinary(process.env.PATH);
+
+// approval-mode maps cleanly onto the normalized 3-tier permission model.
 const PERMISSION_MAP: Record<NormalizedPermission, string> = {
-  readonly: 'read-only',
-  edit: 'workspace',
-  danger: 'danger',
+  readonly: 'always-ask',
+  edit: 'write',
+  danger: 'yolo',
 };
-
 function extractAmpText(o: Record<string, unknown>): string | undefined {
   if (typeof o.text === 'string') return o.text;
   if (typeof o.output === 'string') return o.output;
   if (typeof o.delta === 'string') return o.delta;
   if (typeof o.content === 'string') return o.content;
+  if (isRecord(o.part) && typeof o.part.text === 'string') return o.part.text;
   return undefined;
 }
 
+interface AmpHarnessState {
+  sessionId?: string;
+  costAccum?: number;
+  inputAccum?: number;
+  outputAccum?: number;
+  cacheReadAccum?: number;
+  cacheWriteAccum?: number;
+  turnCount?: number;
+}
+
+function harnessState(state: ParseState): AmpHarnessState {
+  const s = (state._harness ?? {}) as AmpHarnessState;
+  state._harness = s as unknown as Record<string, unknown>;
+  return s;
+}
+
 export function parseAmpLine(line: string, state: ParseState): ParseOutcome {
   let o: unknown;
   try {
     o = JSON.parse(line);
   } catch {
-    if (line.trim().length > 0) return { streamedText: line + '\n', activities: [] };
+    if (line.trim().length > 0) return { streamedText: `${line}\n`, activities: [] };
     return { activities: [] };
   }
   if (!isRecord(o)) return { activities: [] };
   const activities: ParseOutcome['activities'] = [];
   let streamedText: string | undefined;
   const typeStr = typeof o.type === 'string' ? o.type : '';
+  const hs = harnessState(state);
+  // latch session id
+  if (typeStr === 'session' && typeof o.id === 'string') hs.sessionId = o.id;
+  if (isRecord(o.part) && typeof (o.part as Record<string, unknown>).sessionID === 'string')
+    hs.sessionId = (o.part as Record<string, unknown>).sessionID as string;
+  if (typeStr === 'session' && typeof (o as Record<string, unknown>).sessionID === 'string')
+    hs.sessionId = (o as Record<string, unknown>).sessionID as string;
 
-  if (typeStr.includes('tool') || o.type === 'tool_use') {
-    const name = typeof o.name === 'string' ? o.name : 'tool';
-    if (typeStr.includes('start') || o.type === 'tool_use') {
+  // real tool schema: top-level tool_execution_start/tool_execution_end, correlated by toolCallId
+  if (typeStr === 'tool_execution_start' || typeStr === 'tool_execution_end') {
+    const name = typeof o.toolName === 'string' ? o.toolName : 'tool';
+    const id = typeof o.toolCallId === 'string' ? o.toolCallId : undefined;
+    if (typeStr === 'tool_execution_start') {
       activities.push({ kind: 'tool_start', name });
-      if (isRecord(o.input)) activities.push({ kind: 'tool_input', name, input: o.input as Record<string, unknown> });
+      activities.push({
+        kind: 'tool_input',
+        name,
+        input: isRecord(o.args) ? (o.args as Record<string, unknown>) : {},
+        id,
+      });
     } else {
-      activities.push({ kind: 'tool_result', isError: o.is_error === true });
+      activities.push({ kind: 'tool_result', isError: o.isError === true, id });
+    }
+  }
+  if (typeStr === 'message_update' && isRecord(o.assistantMessageEvent)) {
+    const ev = o.assistantMessageEvent as Record<string, unknown>;
+    if (ev.type === 'thinking_delta' && typeof ev.delta === 'string')
+      activities.push({ kind: 'thinking', chars: ev.delta.length });
+    else if (ev.type === 'text_delta' && typeof ev.delta === 'string') streamedText = ev.delta;
+    else if (ev.type === 'thinking_start') activities.push({ kind: 'thinking', chars: 5 });
+  }
+  if (typeStr === 'turn_end' || typeStr === 'agent_end') {
+    const msg = isRecord(o.message)
+      ? (o.message as Record<string, unknown>)
+      : Array.isArray(o.messages) && isRecord(o.messages[o.messages.length - 1])
+        ? (o.messages[o.messages.length - 1] as Record<string, unknown>)
+        : null;
+    if (msg && Array.isArray(msg.content)) {
+      for (const block of msg.content as unknown[]) {
+        if (isRecord(block) && block.type === 'text' && typeof block.text === 'string') streamedText = block.text;
+        if (isRecord(block) && block.type === 'thinking' && typeof block.thinking === 'string')
+          activities.push({ kind: 'thinking', chars: (block.thinking as string).length });
+      }
+    }
+    // real usage lives at message.usage on turn_end (per-turn, not cumulative — see accumulation below)
+    if (typeStr === 'turn_end' && isRecord(msg?.usage)) {
+      const u = msg.usage as Record<string, unknown>;
+      const cost =
+        isRecord(u.cost) && typeof (u.cost as Record<string, unknown>).total === 'number'
+          ? ((u.cost as Record<string, unknown>).total as number)
+          : 0;
+      hs.costAccum = (hs.costAccum ?? 0) + cost;
+      hs.inputAccum = (hs.inputAccum ?? 0) + (typeof u.input === 'number' ? u.input : 0);
+      hs.outputAccum = (hs.outputAccum ?? 0) + (typeof u.output === 'number' ? u.output : 0);
+      hs.cacheReadAccum = (hs.cacheReadAccum ?? 0) + (typeof u.cacheRead === 'number' ? u.cacheRead : 0);
+      hs.cacheWriteAccum = (hs.cacheWriteAccum ?? 0) + (typeof u.cacheWrite === 'number' ? u.cacheWrite : 0);
+      hs.turnCount = (hs.turnCount ?? 0) + 1;
     }
   }
-  if (typeStr.includes('thinking')) activities.push({ kind: 'thinking', chars: 10 });
-
   const text = extractAmpText(o);
-  if (text && !typeStr.includes('tool')) streamedText = text;
-
-  if (o.type === 'result' || o.type === 'done' || o.type === 'completed') {
-    const usage = isRecord(o.usage) ? o.usage : null;
+  if (text && typeStr !== 'tool_execution_start' && typeStr !== 'tool_execution_end' && !streamedText)
+    streamedText = text;
+  if (typeStr === 'turn_end' || typeStr === 'agent_end') {
+    const msg = isRecord(o.message)
+      ? (o.message as Record<string, unknown>)
+      : Array.isArray(o.messages) && isRecord(o.messages[o.messages.length - 1])
+        ? (o.messages[o.messages.length - 1] as Record<string, unknown>)
+        : null;
+    const measured = (hs.turnCount ?? 0) > 0;
     const result: StreamedResult = {
-      result: typeof o.result === 'string' ? o.result : state.streamedText + (streamedText ?? ''),
+      result:
+        typeof o.result === 'string'
+          ? o.result
+          : streamedText
+            ? state.streamedText + streamedText
+            : state.streamedText || (text ?? ''),
       isError: o.is_error === true,
-      numTurns: typeof o.num_turns === 'number' ? o.num_turns : 0,
-      totalCostUsd: typeof o.total_cost_usd === 'number' ? o.total_cost_usd : 0,
-      sessionId: typeof o.session_id === 'string' ? o.session_id : null,
-      stopReason: typeof o.stop_reason === 'string' ? o.stop_reason : null,
+      numTurns: measured ? (hs.turnCount as number) : null,
+      totalCostUsd: measured ? (hs.costAccum as number) : null,
+      sessionId:
+        typeof o.session_id === 'string' ? o.session_id : typeof o.id === 'string' ? o.id : (hs.sessionId ?? null),
+      stopReason:
+        typeof o.stop_reason === 'string'
+          ? o.stop_reason
+          : typeof msg?.stopReason === 'string'
+            ? (msg.stopReason as string)
+            : null,
       permissionDenials: [],
-      durationMs: typeof o.duration_ms === 'number' ? o.duration_ms : null,
+      durationMs:
+        typeof o.duration_ms === 'number'
+          ? o.duration_ms
+          : typeof (o as Record<string, unknown>).duration === 'number'
+            ? ((o as Record<string, unknown>).duration as number)
+            : null,
       durationApiMs: null,
-      ttftMs: null,
-      model: typeof o.model === 'string' ? o.model : null,
+      ttftMs:
+        typeof (o as Record<string, unknown>).ttft === 'number'
+          ? ((o as Record<string, unknown>).ttft as number)
+          : null,
+      model: typeof o.model === 'string' ? o.model : typeof msg?.model === 'string' ? (msg.model as string) : null,
       contextWindow: null,
       maxOutputTokens: null,
-      usage: usage
+      usage: measured
         ? {
-            inputTokens: typeof usage.input_tokens === 'number' ? usage.input_tokens : 0,
-            outputTokens: typeof usage.output_tokens === 'number' ? usage.output_tokens : 0,
-            cacheCreationInputTokens: 0,
-            cacheReadInputTokens: 0,
+            inputTokens: hs.inputAccum ?? 0,
+            outputTokens: hs.outputAccum ?? 0,
+            cacheCreationInputTokens: hs.cacheWriteAccum ?? 0,
+            cacheReadInputTokens: hs.cacheReadAccum ?? 0,
           }
         : null,
     };
+    if (!result.result) result.result = state.streamedText + (streamedText ?? '');
     return { activities, streamedText, result };
   }
   return { activities, streamedText };
 }
-
 export const ampHarness: Harness = {
   name: 'amp',
   displayName: 'Amp',
-  binary: 'amp',
+  binary: RESOLVED_BINARY,
   async detect() {
     try {
       const { stdout } = await execFileAsync('amp', ['--version'], { timeout: 5000 });
@@ -104,11 +224,12 @@ export const ampHarness: Harness = {
     }
   },
   buildArgs(opts: BuildArgsOpts): string[] {
-    const perm = opts.nativePermission ?? PERMISSION_MAP[opts.permission] ?? 'workspace';
-    const args = ['--output', 'jsonl', opts.prompt, '--permission', perm];
+    const approvalMode = opts.nativePermission ?? PERMISSION_MAP[opts.permission] ?? 'always-ask';
+    const args = ['-p', '--mode', 'json', '--approval-mode', approvalMode];
     if (opts.model) args.push('--model', opts.model);
     if (opts.resumeSessionId) args.push('--resume', opts.resumeSessionId);
     for (const dir of opts.addDirs ?? []) args.push('--add-dir', dir);
+    args.push(opts.prompt);
     return args;
   },
   parseLine(line: string, state: ParseState): ParseOutcome {
@@ -117,12 +238,13 @@ export const ampHarness: Harness = {
   extractResult(state: ParseState): StreamedResult | null {
     if (state.result) return state.result;
     if (state.streamedText.trim().length > 0) {
+      const latched = (state._harness as AmpHarnessState | undefined)?.sessionId;
       return {
         result: state.streamedText,
         isError: false,
-        numTurns: 1,
-        totalCostUsd: 0,
-        sessionId: null,
+        numTurns: null,
+        totalCostUsd: null,
+        sessionId: latched ?? null,
         stopReason: null,
         permissionDenials: [],
         durationMs: null,
@@ -136,9 +258,5 @@ export const ampHarness: Harness = {
     }
     return null;
   },
-  permissionMap: {
-    readonly: ['read-only'],
-    edit: ['workspace'],
-    danger: ['danger'],
-  },
+  permissionMap: { readonly: ['always-ask'], edit: ['write'], danger: ['yolo'] },
 };

package/extensions/harnesses/claude.ts

@@ -56,13 +56,18 @@ export function parseClaudeLine(line: string, state: ParseState): ParseOutcome {
           kind: 'tool_input',
           name: block.name,
           input: isRecord(block.input) ? block.input : {},
+          id: typeof block.id === 'string' ? block.id : undefined,
         });
       }
     }
   } else if (o.type === 'user' && isRecord(o.message)) {
     for (const block of Array.isArray(o.message.content) ? o.message.content : []) {
       if (isRecord(block) && block.type === 'tool_result') {
-        activities.push({ kind: 'tool_result', isError: block.is_error === true });
+        activities.push({
+          kind: 'tool_result',
+          isError: block.is_error === true,
+          id: typeof block.tool_use_id === 'string' ? block.tool_use_id : undefined,
+        });
       }
     }
   } else if (o.type === 'result') {
@@ -82,8 +87,8 @@ export function parseClaudeLine(line: string, state: ParseState): ParseOutcome {
     const result: StreamedResult = {
       result: typeof o.result === 'string' ? o.result : state.streamedText,
       isError: o.is_error === true,
-      numTurns: typeof o.num_turns === 'number' ? o.num_turns : 0,
-      totalCostUsd: typeof o.total_cost_usd === 'number' ? o.total_cost_usd : 0,
+      numTurns: typeof o.num_turns === 'number' ? o.num_turns : null,
+      totalCostUsd: typeof o.total_cost_usd === 'number' ? o.total_cost_usd : null,
       sessionId: typeof o.session_id === 'string' ? o.session_id : null,
       stopReason: typeof o.stop_reason === 'string' ? o.stop_reason : null,
       permissionDenials: Array.isArray(o.permission_denials) ? o.permission_denials : [],

package/extensions/harnesses/codex.ts

@@ -9,24 +9,22 @@ import type {
   StreamedResult,
 } from './types.ts';
 
-const execFileAsync = promisify(execFile);
+// Schema verified against codex-cli 0.149.1 — see tests/fixtures/codex.jsonl.
+// `codex exec` in this version dropped `--ask-for-approval` entirely (exec is inherently
+// non-interactive; sandbox alone governs what's allowed) and resume is a subcommand
+// (`exec resume <id> <prompt>`), not a `--thread-id` flag — both confirmed via `codex exec
+// --help` / `codex exec resume --help`, not just the JSONL capture.
 
+const execFileAsync = promisify(execFile);
 function isRecord(v: unknown): v is Record<string, unknown> {
   return typeof v === 'object' && v !== null && !Array.isArray(v);
 }
-
 const SANDBOX_MAP: Record<NormalizedPermission, string> = {
   readonly: 'read-only',
   edit: 'workspace-write',
   danger: 'danger-full-access',
 };
-
-function extractTextFromCodexEvent(o: Record<string, unknown>, state: ParseState): string | undefined {
-  // Codex JSON variants: try common shapes
-  // 1. {type:"item.completed", item:{type:"agent_message", text:"..."}}
-  // 2. {type:"thread.item.completed", item:{type:"agent_message", ...}}
-  // 3. {type:"event", event:{type:"response.output_text.delta", delta:"..."}}
-  // 4. Plain {"text":"..."} or {"output":"..."}
+function extractTextFromCodexEvent(o: Record<string, unknown>, _state: ParseState): string | undefined {
   if (typeof o.text === 'string' && o.type !== 'tool_use') return o.text;
   if (typeof o.output === 'string') return o.output;
   if (typeof o.delta === 'string') return o.delta;
@@ -35,32 +33,102 @@ function extractTextFromCodexEvent(o: Record<string, unknown>, state: ParseState
     return (o.event as Record<string, unknown>).delta as string;
   if (isRecord(o.item) && typeof o.item.text === 'string') return o.item.text;
   if (isRecord(o.item) && typeof o.item.output === 'string') return o.item.output;
-  // agent_message
   if (o.type === 'agent_message' && typeof o.text === 'string') return o.text;
+  if (o.type === 'error' && typeof o.message === 'string') return o.message;
+  if (typeof o.error === 'string') return o.error;
   return undefined;
 }
 
+interface CodexHarnessState {
+  sessionId?: string;
+  turnCount?: number;
+}
+
+function harnessState(state: ParseState): CodexHarnessState {
+  const s = (state._harness ?? {}) as CodexHarnessState;
+  state._harness = s as unknown as Record<string, unknown>;
+  return s;
+}
+
 export function parseCodexLine(line: string, state: ParseState): ParseOutcome {
   let o: unknown;
   try {
     o = JSON.parse(line);
   } catch {
-    // Non-JSON line: treat as streamed text (codex may emit plain text)
-    if (line.trim().length > 0) return { streamedText: line + '\n', activities: [] };
+    if (line.trim().length > 0) return { streamedText: `${line}\n`, activities: [] };
     return { activities: [] };
   }
   if (!isRecord(o)) return { activities: [] };
-
   const activities: ParseOutcome['activities'] = [];
   let streamedText: string | undefined;
-
-  // Tool activity heuristics
-  // Codex may emit: {type:"item.started", item:{type:"tool_use", name:"...", input:{}}}
-  // or {type:"tool_use", name:"..."}
-  const item = isRecord(o.item) ? o.item : null;
   const typeStr = typeof o.type === 'string' ? o.type : '';
-
-  if (typeStr.includes('tool') || typeStr.includes('item.started') || typeStr.includes('function_call')) {
+  const item = isRecord(o.item) ? o.item : null;
+  const hs = harnessState(state);
+  // latch thread_id from thread.started
+  if (typeStr === 'thread.started' && typeof o.thread_id === 'string') {
+    hs.sessionId = o.thread_id;
+    return { activities, streamedText };
+  }
+  if (typeStr === 'turn.started') {
+    hs.turnCount = (hs.turnCount ?? 0) + 1;
+    return { activities, streamedText };
+  }
+  if (typeStr === 'error' && typeof o.message === 'string') {
+    streamedText = o.message;
+  }
+  if (typeof o.error === 'string' && !streamedText) streamedText = o.error;
+  if (typeStr === 'turn.failed') {
+    const msg =
+      isRecord(o.error) && typeof o.error.message === 'string'
+        ? o.error.message
+        : typeof o.error === 'string'
+          ? o.error
+          : typeof o.message === 'string'
+            ? o.message
+            : state.streamedText + (streamedText ?? '');
+    const result: StreamedResult = {
+      result: msg,
+      isError: true,
+      numTurns: null,
+      totalCostUsd: null,
+      sessionId: typeof o.thread_id === 'string' ? o.thread_id : (hs.sessionId ?? null),
+      stopReason: 'error',
+      permissionDenials: [],
+      durationMs: null,
+      durationApiMs: null,
+      ttftMs: null,
+      model: null,
+      contextWindow: null,
+      maxOutputTokens: null,
+      usage: null,
+    };
+    return { activities, streamedText, result };
+  }
+  // real tool schema: item.started/item.completed carry item.id (correlating id) and item.type
+  // (no separate "name" field — command_execution is the only item type observed emitting a
+  // shell command; other item types like file_change/mcp_tool_call may exist but weren't
+  // captured, so the generic name/input guesses below stay as a fallback for those).
+  if (
+    item &&
+    (typeStr === 'item.started' || typeStr === 'item.completed') &&
+    item.type !== 'agent_message' &&
+    item.type !== 'reasoning'
+  ) {
+    const id = typeof item.id === 'string' ? item.id : undefined;
+    const name = typeof item.name === 'string' ? item.name : typeof item.type === 'string' ? item.type : 'tool';
+    if (typeStr === 'item.started') {
+      const input =
+        typeof item.command === 'string' ? { command: item.command } : isRecord(item.input) ? item.input : {};
+      activities.push({ kind: 'tool_start', name });
+      activities.push({ kind: 'tool_input', name, input: input as Record<string, unknown>, id });
+    } else {
+      const isError =
+        (typeof item.exit_code === 'number' && item.exit_code !== 0) ||
+        item.status === 'failed' ||
+        item.is_error === true;
+      activities.push({ kind: 'tool_result', isError, id });
+    }
+  } else if (typeStr.includes('tool') || typeStr.includes('function_call')) {
     const toolName = (item && typeof item.name === 'string' ? item.name : typeof o.name === 'string' ? o.name : null) as
       | string
       | null;
@@ -68,37 +136,37 @@ export function parseCodexLine(line: string, state: ParseState): ParseOutcome {
       if (typeStr.includes('started') || o.type === 'tool_use') {
         activities.push({ kind: 'tool_start', name: toolName });
         if (item && isRecord(item.input))
-          activities.push({
-            kind: 'tool_input',
-            name: toolName,
-            input: item.input as Record<string, unknown>,
-          });
+          activities.push({ kind: 'tool_input', name: toolName, input: item.input as Record<string, unknown> });
         else if (isRecord(o.input))
           activities.push({ kind: 'tool_input', name: toolName, input: o.input as Record<string, unknown> });
-      } else if (typeStr.includes('completed') || typeStr.includes('result')) {
+      } else if (typeStr.includes('completed') || typeStr.includes('result'))
         activities.push({ kind: 'tool_result', isError: o.is_error === true || o.error === true });
-      }
     }
   }
-  if (o.type === 'tool_result' || (typeStr === 'item.completed' && item?.type === 'tool_result')) {
+  if (o.type === 'tool_result')
     activities.push({ kind: 'tool_result', isError: (o as Record<string, unknown>).is_error === true });
-  }
   if (typeStr.includes('thinking') || o.type === 'reasoning' || item?.type === 'reasoning') {
     const thinkingText = extractTextFromCodexEvent(o, state);
     if (thinkingText) activities.push({ kind: 'thinking', chars: thinkingText.length });
     else activities.push({ kind: 'thinking', chars: 10 });
   }
-
-  // Text extraction
   const text = extractTextFromCodexEvent(o, state);
-  if (text && !typeStr.includes('tool') && !typeStr.includes('thinking')) {
-    // Avoid double-counting tool inputs as text
+  if (
+    text &&
+    !typeStr.includes('tool') &&
+    !typeStr.includes('thinking') &&
+    typeStr !== 'error' &&
+    typeStr !== 'turn.failed'
+  )
     streamedText = text;
-  }
-
-  // Result detection: final result often {type:"result", result:"...", total_cost_usd, usage, ...}
-  // or {type:"thread.completed", ...}
-  if (o.type === 'result' || o.type === 'thread.completed' || o.type === 'task.completed') {
+  // real final-usage event is turn.completed (thread.completed/task.completed/result kept as
+  // fallback in case another codex-cli version emits them instead).
+  if (
+    o.type === 'result' ||
+    o.type === 'thread.completed' ||
+    o.type === 'task.completed' ||
+    typeStr === 'turn.completed'
+  ) {
     const usage = isRecord(o.usage) ? o.usage : null;
     const result: StreamedResult = {
       result:
@@ -108,8 +176,16 @@ export function parseCodexLine(line: string, state: ParseState): ParseOutcome {
             ? o.output
             : state.streamedText + (streamedText ?? ''),
       isError: o.is_error === true || o.error === true,
-      numTurns: typeof o.num_turns === 'number' ? o.num_turns : typeof o.turns === 'number' ? o.turns : 0,
-      totalCostUsd: typeof o.total_cost_usd === 'number' ? o.total_cost_usd : typeof o.cost === 'number' ? o.cost : 0,
+      numTurns:
+        typeof o.num_turns === 'number'
+          ? o.num_turns
+          : typeof o.turns === 'number'
+            ? o.turns
+            : hs.turnCount
+              ? hs.turnCount
+              : null,
+      totalCostUsd:
+        typeof o.total_cost_usd === 'number' ? o.total_cost_usd : typeof o.cost === 'number' ? o.cost : null,
       sessionId:
         typeof o.session_id === 'string'
           ? o.session_id
@@ -117,7 +193,7 @@ export function parseCodexLine(line: string, state: ParseState): ParseOutcome {
             ? o.thread_id
             : typeof o.id === 'string'
               ? o.id
-              : null,
+              : (hs.sessionId ?? null),
       stopReason: typeof o.stop_reason === 'string' ? o.stop_reason : null,
       permissionDenials: Array.isArray(o.permission_denials) ? o.permission_denials : [],
       durationMs: typeof o.duration_ms === 'number' ? o.duration_ms : null,
@@ -128,35 +204,29 @@ export function parseCodexLine(line: string, state: ParseState): ParseOutcome {
       maxOutputTokens: null,
       usage: usage
         ? {
-            inputTokens:
-              typeof usage.input_tokens === 'number'
-                ? usage.input_tokens
-                : typeof (usage as Record<string, unknown>).inputTokens === 'number'
-                  ? ((usage as Record<string, unknown>).inputTokens as number)
-                  : 0,
-            outputTokens:
-              typeof usage.output_tokens === 'number'
-                ? usage.output_tokens
-                : typeof (usage as Record<string, unknown>).outputTokens === 'number'
-                  ? ((usage as Record<string, unknown>).outputTokens as number)
-                  : 0,
+            // real fields: input_tokens, cached_input_tokens, cache_write_input_tokens, output_tokens
+            // (no total_cost_usd anywhere in this schema — ChatGPT-plan auth doesn't report $ cost).
+            inputTokens: typeof usage.input_tokens === 'number' ? usage.input_tokens : 0,
+            outputTokens: typeof usage.output_tokens === 'number' ? usage.output_tokens : 0,
             cacheCreationInputTokens:
-              typeof (usage as Record<string, unknown>).cache_creation_input_tokens === 'number'
-                ? ((usage as Record<string, unknown>).cache_creation_input_tokens as number)
-                : 0,
+              typeof usage.cache_write_input_tokens === 'number'
+                ? usage.cache_write_input_tokens
+                : typeof (usage as Record<string, unknown>).cache_creation_input_tokens === 'number'
+                  ? ((usage as Record<string, unknown>).cache_creation_input_tokens as number)
+                  : 0,
             cacheReadInputTokens:
-              typeof (usage as Record<string, unknown>).cache_read_input_tokens === 'number'
-                ? ((usage as Record<string, unknown>).cache_read_input_tokens as number)
-                : 0,
+              typeof usage.cached_input_tokens === 'number'
+                ? usage.cached_input_tokens
+                : typeof (usage as Record<string, unknown>).cache_read_input_tokens === 'number'
+                  ? ((usage as Record<string, unknown>).cache_read_input_tokens as number)
+                  : 0,
           }
         : null,
     };
     return { activities, streamedText, result };
   }
-
   return { activities, streamedText };
 }
-
 export const codexHarness: Harness = {
   name: 'codex',
   displayName: 'Muse',
@@ -170,15 +240,11 @@ export const codexHarness: Harness = {
     }
   },
   buildArgs(opts: BuildArgsOpts): string[] {
-    // codex exec --json <prompt> --sandbox <level> --ask-for-approval <level>
     const sandbox = opts.nativePermission ?? SANDBOX_MAP[opts.permission] ?? 'workspace-write';
-    const args = ['exec', '--json', opts.prompt, '--sandbox', sandbox];
-    if (opts.permission === 'danger' || sandbox === 'danger-full-access') args.push('--ask-for-approval', 'never');
-    else if (opts.permission === 'readonly') args.push('--ask-for-approval', 'never');
-    else args.push('--ask-for-approval', 'on-request');
+    const args = opts.resumeSessionId
+      ? ['exec', 'resume', opts.resumeSessionId, opts.prompt, '--json']
+      : ['exec', '--json', opts.prompt, '--sandbox', sandbox];
     if (opts.model) args.push('--model', opts.model);
-    if (opts.resumeSessionId) args.push('--thread-id', opts.resumeSessionId);
-    // maxBudgetUsd not natively supported; pass as env hint via --config if needed
     for (const dir of opts.addDirs ?? []) args.push('--add-dir', dir);
     return args;
   },
@@ -187,14 +253,14 @@ export const codexHarness: Harness = {
   },
   extractResult(state: ParseState): StreamedResult | null {
     if (state.result) return state.result;
-    // Fallback: if no explicit result, synthesize from streamed text if any
     if (state.streamedText.trim().length > 0) {
+      const latched = (state._harness as CodexHarnessState | undefined)?.sessionId;
       return {
         result: state.streamedText,
         isError: false,
-        numTurns: 1,
-        totalCostUsd: 0,
-        sessionId: null,
+        numTurns: null,
+        totalCostUsd: null,
+        sessionId: latched ?? null,
         stopReason: null,
         permissionDenials: [],
         durationMs: null,
@@ -208,9 +274,5 @@ export const codexHarness: Harness = {
     }
     return null;
   },
-  permissionMap: {
-    readonly: ['read-only'],
-    edit: ['workspace-write'],
-    danger: ['danger-full-access'],
-  },
+  permissionMap: { readonly: ['read-only'], edit: ['workspace-write'], danger: ['danger-full-access'] },
 };