From 91ad64e3329fcf661fd670d6f75bde234389c6b6 Mon Sep 17 00:00:00 2001 From: lisihao Date: Fri, 5 Jun 2026 21:50:39 -0400 Subject: [PATCH] Add AI influence scratch chat reuse safeguards --- harness/scripts/tech_hotspot_radar.py | 356 +++++++++++++++++- .../test_tech_hotspot_browser_agent_cmd.py | 263 +++++++++++++ 2 files changed, 600 insertions(+), 19 deletions(-) diff --git a/harness/scripts/tech_hotspot_radar.py b/harness/scripts/tech_hotspot_radar.py index 3d22d103e..47d67b297 100755 --- a/harness/scripts/tech_hotspot_radar.py +++ b/harness/scripts/tech_hotspot_radar.py @@ -13822,7 +13822,7 @@ def build_ai_influence_video_grouping_prompt(materials: list[dict[str, Any]], *, } for item in materials ] - return f"""你是 AI Influence 的 YouTube 素材总编和研究策展人。 + body = f"""你是 AI Influence 的 YouTube 素材总编和研究策展人。 你现在使用 Browser Agent 算子打开 ChatGPT，模型必须是 {model_name}，Thinking high。 @@ -13881,6 +13881,20 @@ def build_ai_influence_video_grouping_prompt(materials: list[dict[str, Any]], *, 视频 transcript 材料 JSON： {json.dumps(safe_materials, ensure_ascii=False, indent=2)} """ + return wrap_ai_influence_scratch_chat_prompt( + body, + lane="ai_influence_planned_report", + task_type="grouping", + run_id=f"ai-influence-grouping:{date_str}:{days}d", + report_date=date_str, + extra_guard=( + "[GROUPING_SCOPE_GUARD]\n" + "你当前只做素材语义分组，不做报告规划，不做章节写作。\n" + "禁止沿用历史对话中的旧 group、旧趋势桶、旧事件归类；除非本次 payload 再次给出。\n" + "如果当前 transcript 材料不足，就输出 weak_misc / ungrouped_materials，不要用历史聊天补完。\n" + "[/GROUPING_SCOPE_GUARD]" + ), + ) def normalize_ai_influence_video_groups(group_plan: dict[str, Any], catalog: list[dict[str, Any]]) -> dict[str, Any]: @@ -13940,7 +13954,7 @@ def build_ai_influence_report_plan_prompt(catalog: list[dict[str, Any]], *, for item in catalog ] group_plan = video_group_plan or {"video_groups": [], "planning_guidance": []} - return f"""你是 AI Influence 的总编辑、研究主编和技术趋势报告规划师。 + body = f"""你是 AI Influence 的总编辑、研究主编和技术趋势报告规划师。 你现在使用的是 Browser Agent 算子打开 ChatGPT，模型必须是 {model_name}，Thinking high。 @@ -14042,6 +14056,78 @@ def build_ai_influence_report_plan_prompt(catalog: list[dict[str, Any]], *, 视频目录 JSON： {json.dumps(safe_catalog, ensure_ascii=False, indent=2)} """ + return wrap_ai_influence_scratch_chat_prompt( + body, + lane="ai_influence_planned_report", + task_type="planner", + run_id=f"ai-influence-planner:{date_str}:{days}d", + report_date=date_str, + extra_guard=( + "[PLANNER_SCOPE_GUARD]\n" + "你现在不是在续写旧报告，也不是在修订之前的 report plan。\n" + "你只能基于当前消息提供的 `video_group_plan` 和 `video catalog` 重新规划。\n" + "禁止沿用任何历史报告标题、趋势结构、章节树、figure slots，除非它们在当前 payload 中再次明确给出。\n" + "如果当前素材不足以支撑多份报告，就减少 reports 数量；不要因为历史习惯维持旧数量。\n" + "[/PLANNER_SCOPE_GUARD]" + ), + ) + + +def _ai_influence_scratch_identity(*, + lane: str, + task_type: str, + run_id: str, + report_date: str, + report_id: str | None = None, + chapter_id: str | None = None) -> str: + return "\n".join([ + "当前任务身份：", + f"- lane: {lane}", + f"- task_type: {task_type}", + f"- run_id: {run_id}", + f"- report_date: {report_date}", + f"- report_id: {str(report_id or 'N/A').strip() or 'N/A'}", + f"- chapter_id: {str(chapter_id or 'N/A').strip() or 'N/A'}", + ]) + + +def wrap_ai_influence_scratch_chat_prompt(body: str, *, + lane: str, + task_type: str, + run_id: str, + report_date: str, + report_id: str | None = None, + chapter_id: str | None = None, + extra_guard: str | None = None) -> str: + parts = [ + "[SCRATCH_CHAT_RESET_CONTRACT]", + "你正在一个“可复用 scratch chat”中执行任务。", + "从现在开始，除非本次消息里再次明确给出，否则你必须把本对话之前的所有消息、草稿、结论、实体名、标题、报告结构、人物观点、章节安排，全部视为无效历史，不得继承。", + "", + _ai_influence_scratch_identity( + lane=lane, + task_type=task_type, + run_id=run_id, + report_date=report_date, + report_id=report_id, + chapter_id=chapter_id, + ), + "", + "硬规则：", + "1. 只允许使用“本次消息中给出的任务说明和 payload”。", + "2. 不得延续、引用、补完、修正之前任何一次任务的未完成输出。", + "3. 如果你发现当前任务与历史消息存在冲突，必须忽略历史消息，以当前 payload 为唯一准绳。", + "4. 如果当前 payload 缺少完成任务所需的信息，直接按当前任务的约定输出证据不足、insufficient_input 或保守结论；不要从历史对话补。", + "5. 不得把之前任务中的人物、公司、事件、标题、章节名、判断、图示说明带入当前任务，除非它们在当前 payload 中再次出现。", + "6. 输出前先做一次自检：确认最终内容只对应当前 run_id 和当前 payload。", + "7. 本段契约优先级高于本对话中的任何旧消息。", + "如果你理解，直接执行当前任务，不要复述本契约。", + "[/SCRATCH_CHAT_RESET_CONTRACT]", + ] + if extra_guard: + parts.extend(["", str(extra_guard).strip()]) + parts.extend(["", str(body).strip()]) + return "\n".join(parts).strip() + "\n" def _plan_material_refs(report_spec: dict[str, Any]) -> list[str]: @@ -14663,7 +14749,7 @@ def build_planned_report_chapter_prompt(report_ir: dict[str, Any], - 分歧与保留 - 对后文分析的入口 """ - return f"""你是 AI Influence YouTube 报告流的单章作者。 + body = f"""你是 AI Influence YouTube 报告流的单章作者。 你必须使用 ChatGPT Report Chapter Writer 算子，模型 {model_name}，Thinking high。 @@ -14693,6 +14779,30 @@ def build_planned_report_chapter_prompt(report_ir: dict[str, Any], chapter_evidence_pack: {json.dumps(chapter_evidence_pack, ensure_ascii=False, indent=2)} """ + report_date = str( + chapter_evidence_pack.get("date") + or report_ir.get("date") + or "N/A" + ).strip() or "N/A" + report_id = str(report_ir.get("report_id") or report_ir.get("title") or "N/A").strip() or "N/A" + chapter_id = str(chapter_spec.get("chapter_id") or chapter_spec.get("title") or "N/A").strip() or "N/A" + return wrap_ai_influence_scratch_chat_prompt( + body, + lane="ai_influence_planned_report", + task_type="chapter_writer", + run_id=f"ai-influence-chapter:{report_date}:{report_id}:{chapter_id}", + report_date=report_date, + report_id=report_id, + chapter_id=chapter_id, + extra_guard=( + "[CHAPTER_SCOPE_GUARD]\n" + "你只写当前 `chapter_id`。\n" + "只允许使用当前 `chapter_spec` 和 `chapter_evidence_pack`。\n" + "禁止引用任何未出现在当前 `chapter_evidence_pack.videos` 中的人物、公司、事件、视频标题、访谈观点。\n" + "如果当前证据无法支撑强判断，就明确写“证据不足 / 暂不作为强结论”，不要从历史对话补证据。\n" + "[/CHAPTER_SCOPE_GUARD]" + ), + ) def synthesize_ai_influence_chapter_report(report_ir: dict[str, Any], @@ -14735,6 +14845,7 @@ def call_ai_influence_chapter_writer_with_repair(chapter_prompt: str, purpose: str, model_name: str, chapter_id: str, + scratch_profile_id: str | None = None, min_chars: int = 120, max_attempts: int = 3) -> dict[str, Any]: """Run ChatGPT Report Chapter Writer with a bounded repair loop. @@ -14764,6 +14875,11 @@ def call_ai_influence_chapter_writer_with_repair(chapter_prompt: str, purpose=purpose if attempt == 1 else f"{purpose}-repair-{attempt}", requested_model=model_name, operator_kind="chapter_writer", + scratch_profile_id=scratch_profile_id, + open_project_first=False, + require_project=False, + force_new_chat=False, + require_isolated_conversation=False, ) markdown = str(result.get("markdown") or "").strip() if len(markdown) >= min_chars: @@ -14896,15 +15012,110 @@ def build_phase_report_markdown_prompt(evidence_pack: dict[str, Any], model_name def _browser_agent_request_dir(config: dict[str, Any], purpose: str) -> Path: - state_dir = Path((config.get("output") or {}).get( - "state_dir", str(Path.home() / ".solar/harness/state/tech-hotspot-radar") - )).expanduser() + state_dir = _expand_runtime_path( + (config.get("output") or {}).get( + "state_dir", + str(Path.home() / ".solar/harness/state/tech-hotspot-radar"), + ) + ) stamp = dt.datetime.now(UTC).strftime("%Y%m%dT%H%M%SZ") out = state_dir / "browser-agent-requests" / f"{stamp}-{slugify(purpose)[:60]}" out.mkdir(parents=True, exist_ok=True) return out +def _browser_agent_state_dir(config: dict[str, Any]) -> Path: + return _expand_runtime_path( + (config.get("output") or {}).get( + "state_dir", + str(Path.home() / ".solar/harness/state/tech-hotspot-radar"), + ) + ) + + +def _browser_agent_scratch_session_path(config: dict[str, Any], scratch_profile_id: str) -> Path: + normalized = Path(str(scratch_profile_id).strip()) + return _browser_agent_state_dir(config) / "browser-agent-scratch-pool" / normalized / "session.json" + + +def _load_browser_agent_scratch_session(config: dict[str, Any], scratch_profile_id: str) -> dict[str, Any]: + path = _browser_agent_scratch_session_path(config, scratch_profile_id) + if not path.exists(): + return {} + try: + data = json.loads(path.read_text(encoding="utf-8")) + except Exception: + return {} + return data if isinstance(data, dict) else {} + + +def _write_browser_agent_scratch_session(config: dict[str, Any], scratch_profile_id: str, payload: dict[str, Any]) -> None: + path = _browser_agent_scratch_session_path(config, scratch_profile_id) + path.parent.mkdir(parents=True, exist_ok=True) + path.write_text(json.dumps(payload, ensure_ascii=False, indent=2) + "\n", encoding="utf-8") + + +def _conversation_state_from_request_dir(request_dir: Path) -> dict[str, Any]: + candidates = [ + request_dir / "page.json", + request_dir / "conversation.json", + request_dir / "submitted-run.json", + request_dir / "post-submit-state.json", + ] + for path in candidates: + if not path.exists(): + continue + try: + data = json.loads(path.read_text(encoding="utf-8")) + except Exception: + continue + if isinstance(data, dict): + return data + return {} + + +def _persist_browser_agent_scratch_session_from_request_dir( + config: dict[str, Any], + *, + scratch_profile_id: str | None, + purpose: str, + request_dir: Path, +) -> bool: + if not scratch_profile_id: + return False + conversation_state = _conversation_state_from_request_dir(request_dir) + conversation_url = str( + conversation_state.get("url") + or conversation_state.get("conversation_url") + or "" + ).strip() + conversation_id = str(conversation_state.get("conversation_id") or "").strip() + has_conversation_path = "/c/" in conversation_url + if not conversation_url or (not conversation_id and not has_conversation_path): + return False + _write_browser_agent_scratch_session( + config, + scratch_profile_id, + { + "scratch_profile_id": scratch_profile_id, + "purpose": purpose, + "conversation_url": conversation_url, + "conversation_id": conversation_id, + "updated_at": iso_z(), + "request_dir": str(request_dir), + }, + ) + return True + + +def ai_influence_scratch_profile_id(*, task_type: str, report_id: str | None = None) -> str: + clean_task = slugify(str(task_type or "").strip())[:40] or "default" + if clean_task == "chapter_writer": + clean_report = slugify(str(report_id or "").strip())[:80] or "shared" + return f"chatgpt/ai-influence-planned/chapter/{clean_report}" + return f"chatgpt/ai-influence-planned/{clean_task}" + + def browser_agent_chatgpt_cmd(config: dict[str, Any]) -> list[str]: """Resolve the browser-agent ChatGPT executor command. @@ -15003,7 +15214,8 @@ def call_browser_agent_chatgpt_text(prompt: str, config: dict[str, Any], *, open_project_first: bool | None = None, require_project: bool | None = None, force_new_chat: bool | None = None, - require_isolated_conversation: bool | None = None) -> dict[str, Any]: + require_isolated_conversation: bool | None = None, + scratch_profile_id: str | None = None) -> dict[str, Any]: reasoner_cfg = ((config.get("youtube") or {}).get("phase_report_reasoner") or {}) flow_cfg = ((config.get("youtube") or {}).get("ai_influence_report_flow") or {}) writer_cfg = (flow_cfg.get("report_writer") or {}) @@ -15028,6 +15240,7 @@ def call_browser_agent_chatgpt_text(prompt: str, config: dict[str, Any], *, "created_at": iso_z(), "status": "pending_executor", "note": "AI Influence high-judgment stages must use DeepResearchChatGPT/chatgpt_thinking_high via Browser Agent + ChatGPT 5.5 Thinking high. No Codex/local fallback is allowed.", + "scratch_profile_id": str(scratch_profile_id or "").strip(), } (req_dir / "request.json").write_text(json.dumps(meta, ensure_ascii=False, indent=2) + "\n", encoding="utf-8") cmd = browser_agent_chatgpt_cmd(config) @@ -15051,6 +15264,18 @@ def call_browser_agent_chatgpt_text(prompt: str, config: dict[str, Any], *, env["CHATGPT_REPORT_OPERATOR_KIND"] = operator_kind if target_url: env["BROWSER_AGENT_CHATGPT_URL"] = str(target_url) + resumed_target_url = "" + if scratch_profile_id: + env["BROWSER_AGENT_PROFILE_ID"] = str(scratch_profile_id) + scratch_session = _load_browser_agent_scratch_session(config, scratch_profile_id) + resumed_target_url = str( + scratch_session.get("conversation_url") + or scratch_session.get("url") + or "" + ).strip() + if resumed_target_url and not target_url: + env["BROWSER_AGENT_CHATGPT_URL"] = resumed_target_url + meta["scratch_resume_url"] = resumed_target_url resolved_headless = headless if resolved_headless is None: resolved_headless = _env_override_bool("BROWSER_AGENT_HEADLESS", "TECH_HOTSPOT_BROWSER_CHATGPT_HEADLESS") @@ -15155,21 +15380,77 @@ def call_browser_agent_chatgpt_text(prompt: str, config: dict[str, Any], *, if project_name: env["BROWSER_AGENT_CHATGPT_PROJECT_NAME"] = project_name started = time.time() - run = subprocess.run( + proc = subprocess.Popen( cmd, - input=prompt, - text=True, + stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, - timeout=timeout, + text=True, env=env, ) - output = _strip_browser_agent_noise(run.stdout or "") + if proc.stdin is not None: + proc.stdin.write(prompt) + proc.stdin.close() + + scratch_session_persisted = False + timeout_deadline = started + timeout + while True: + if not scratch_session_persisted: + scratch_session_persisted = _persist_browser_agent_scratch_session_from_request_dir( + config, + scratch_profile_id=scratch_profile_id, + purpose=purpose, + request_dir=req_dir, + ) + try: + proc.wait(timeout=0.5) + break + except subprocess.TimeoutExpired: + if time.time() >= timeout_deadline: + if not scratch_session_persisted: + scratch_session_persisted = _persist_browser_agent_scratch_session_from_request_dir( + config, + scratch_profile_id=scratch_profile_id, + purpose=purpose, + request_dir=req_dir, + ) + proc.kill() + output = "" + if proc.stdout is not None: + output = proc.stdout.read() or "" + output = _strip_browser_agent_noise(output) + (req_dir / "stdout.txt").write_text(output + ("\n" if output else ""), encoding="utf-8") + raise subprocess.TimeoutExpired(cmd=cmd, timeout=timeout, output=output) + output = "" + if proc.stdout is not None: + output = proc.stdout.read() or "" + output = _strip_browser_agent_noise(output) (req_dir / "stdout.txt").write_text(output + ("\n" if output else ""), encoding="utf-8") - if run.returncode != 0: - raise RuntimeError(f"browser_agent_chatgpt failed rc={run.returncode}: {output[-2000:]}") + if proc.returncode != 0: + if not scratch_session_persisted: + scratch_session_persisted = _persist_browser_agent_scratch_session_from_request_dir( + config, + scratch_profile_id=scratch_profile_id, + purpose=purpose, + request_dir=req_dir, + ) + raise RuntimeError(f"browser_agent_chatgpt failed rc={proc.returncode}: {output[-2000:]}") if len(output) < (500 if expected == "json" else 1000): + if not scratch_session_persisted: + scratch_session_persisted = _persist_browser_agent_scratch_session_from_request_dir( + config, + scratch_profile_id=scratch_profile_id, + purpose=purpose, + request_dir=req_dir, + ) raise ValueError(f"browser_agent_chatgpt output too short: {len(output)} chars") + if not scratch_session_persisted: + _persist_browser_agent_scratch_session_from_request_dir( + config, + scratch_profile_id=scratch_profile_id, + purpose=purpose, + request_dir=req_dir, + ) meta.update({ "status": "completed", "latency_ms": int((time.time() - started) * 1000), @@ -15251,7 +15532,12 @@ def extract_json_payload_lenient(text: str) -> dict[str, Any]: def call_browser_agent_chatgpt_markdown(prompt: str, config: dict[str, Any], *, purpose: str, requested_model: str | None = None, - operator_kind: str | None = None) -> dict[str, Any]: + operator_kind: str | None = None, + scratch_profile_id: str | None = None, + force_new_chat: bool | None = None, + require_isolated_conversation: bool | None = None, + open_project_first: bool | None = None, + require_project: bool | None = None) -> dict[str, Any]: result = call_browser_agent_chatgpt_text( prompt, config, @@ -15259,6 +15545,11 @@ def call_browser_agent_chatgpt_markdown(prompt: str, config: dict[str, Any], *, expected="markdown", requested_model=requested_model, operator_kind=operator_kind, + scratch_profile_id=scratch_profile_id, + force_new_chat=force_new_chat, + require_isolated_conversation=require_isolated_conversation, + open_project_first=open_project_first, + require_project=require_project, ) markdown = str(result.pop("text") or "").strip() result["markdown"] = markdown @@ -15630,7 +15921,8 @@ def call_browser_agent_chatgpt_json(prompt: str, config: dict[str, Any], *, open_project_first: bool | None = None, require_project: bool | None = None, force_new_chat: bool | None = None, - require_isolated_conversation: bool | None = None) -> dict[str, Any]: + require_isolated_conversation: bool | None = None, + scratch_profile_id: str | None = None) -> dict[str, Any]: result = call_browser_agent_chatgpt_text( prompt, config, @@ -15650,6 +15942,7 @@ def call_browser_agent_chatgpt_json(prompt: str, config: dict[str, Any], *, require_project=require_project, force_new_chat=force_new_chat, require_isolated_conversation=require_isolated_conversation, + scratch_profile_id=scratch_profile_id, ) payload = extract_json_payload_lenient(str(result.pop("text") or "")) payload["_backend"] = result["backend"] @@ -17163,16 +17456,29 @@ def cmd_phase_report(args: argparse.Namespace) -> int: return 1 +def _expand_runtime_path(value: str | Path) -> Path: + raw = str(value) + defaults = { + "${SOLAR_KNOWLEDGE_DIR}": str(Path.home() / "Knowledge"), + "${HARNESS_DIR}": str(Path.home() / ".solar" / "harness"), + "${SOLAR_REPO}": str(Path.home() / "Solar"), + } + for token, replacement in defaults.items(): + if token in raw and token.replace("${", "").replace("}", "") not in os.environ: + raw = raw.replace(token, replacement) + return Path(os.path.expandvars(raw)).expanduser() + + def _ai_influence_planned_base(config: dict[str, Any], args: argparse.Namespace, date_str: str) -> Path: - raw_base = Path( + raw_base = _expand_runtime_path( getattr(args, "output_base", None) or (config.get("output") or {}).get("raw_dir", "${SOLAR_KNOWLEDGE_DIR}/_raw/tech-hotspot-radar") - ).expanduser() + ) return raw_base / "ai-influence-planned" / date_str def _ai_influence_canonical_planned_base(config: dict[str, Any], date_str: str) -> Path: - raw_base = Path((config.get("output") or {}).get("raw_dir", "${SOLAR_KNOWLEDGE_DIR}/_raw/tech-hotspot-radar")).expanduser() + raw_base = _expand_runtime_path((config.get("output") or {}).get("raw_dir", "${SOLAR_KNOWLEDGE_DIR}/_raw/tech-hotspot-radar")) return raw_base / "ai-influence-planned" / date_str @@ -17367,6 +17673,11 @@ def cmd_plan_ai_influence_reports(args: argparse.Namespace) -> int: purpose=f"ai-influence-video-grouping-{date_str}", requested_model=model_name, operator_kind="planner", + scratch_profile_id=ai_influence_scratch_profile_id(task_type="grouping"), + open_project_first=False, + require_project=False, + force_new_chat=False, + require_isolated_conversation=False, ) group_plan = normalize_ai_influence_video_groups(raw_group_plan, catalog) group_plan["catalog_video_count"] = len(catalog) @@ -17408,6 +17719,11 @@ def cmd_plan_ai_influence_reports(args: argparse.Namespace) -> int: purpose=f"ai-influence-report-plan-{date_str}", requested_model=model_name, operator_kind="planner", + scratch_profile_id=ai_influence_scratch_profile_id(task_type="planner"), + open_project_first=False, + require_project=False, + force_new_chat=False, + require_isolated_conversation=False, ) plan["catalog_video_count"] = len(catalog) plan["video_group_count"] = len(group_plan.get("video_groups") or []) @@ -17576,6 +17892,7 @@ def writer_callable(chapter_spec: dict[str, Any], chapter_evidence_pack: dict[st purpose=f"ai-influence-report-chapter-{date_str}-{report_id}-{chapter_id}", model_name=writer_model, chapter_id=chapter_id, + scratch_profile_id=ai_influence_scratch_profile_id(task_type="chapter_writer", report_id=report_id), ) for job in jobs: @@ -17824,6 +18141,7 @@ def _cmd_run_ai_influence_planned_reports_legacy(args: argparse.Namespace) -> in purpose=f"ai-influence-report-chapter-{date_str}-{report_id}-{chapter_id}", model_name=model_name, chapter_id=chapter_id, + scratch_profile_id=ai_influence_scratch_profile_id(task_type="chapter_writer", report_id=report_id), ) chapter_markdown = str(chapter_result.get("markdown") or "").strip() if len(chapter_markdown) < 120: diff --git a/harness/tests/test_tech_hotspot_browser_agent_cmd.py b/harness/tests/test_tech_hotspot_browser_agent_cmd.py index d818e08b5..7796cd727 100644 --- a/harness/tests/test_tech_hotspot_browser_agent_cmd.py +++ b/harness/tests/test_tech_hotspot_browser_agent_cmd.py @@ -3,9 +3,13 @@ import json import os import sqlite3 +import subprocess import sys +import textwrap from pathlib import Path +import pytest + ROOT = Path(__file__).resolve().parents[2] SCRIPT = ROOT / "harness" / "scripts" / "tech_hotspot_radar.py" @@ -952,6 +956,9 @@ def test_grouping_prompt_uses_transcript_and_material_type_contract(): assert "tutorial_demo" in prompt assert "transcript_excerpt" in prompt assert "不要只按关键词或发布时间聚类" in prompt + assert "[SCRATCH_CHAT_RESET_CONTRACT]" in prompt + assert "task_type: grouping" in prompt + assert "[GROUPING_SCOPE_GUARD]" in prompt def test_plan_prompt_consumes_video_groups_and_requires_hierarchy(): @@ -989,6 +996,9 @@ def test_plan_prompt_consumes_video_groups_and_requires_hierarchy(): assert '"trends"' in prompt assert '"subsections"' in prompt assert "趋势 X → 章节 Y → 小结 Z" in prompt + assert "[SCRATCH_CHAT_RESET_CONTRACT]" in prompt + assert "task_type: planner" in prompt + assert "[PLANNER_SCOPE_GUARD]" in prompt def test_plan_material_refs_recurses_trends_chapters_subsections(): @@ -1012,6 +1022,252 @@ def test_plan_material_refs_recurses_trends_chapters_subsections(): assert refs == ["V001", "V002", "V003", "V004", "V005"] +def test_ai_influence_scratch_profile_id_partitions_task_types(): + ns = _load_namespace() + assert ns["ai_influence_scratch_profile_id"](task_type="grouping") == "chatgpt/ai-influence-planned/grouping" + assert ns["ai_influence_scratch_profile_id"](task_type="planner") == "chatgpt/ai-influence-planned/planner" + assert ns["ai_influence_scratch_profile_id"]( + task_type="chapter_writer", + report_id="edge-ai-physical-ai", + ) == "chatgpt/ai-influence-planned/chapter/edge-ai-physical-ai" + + +def test_browser_agent_state_dir_expands_harness_placeholder(): + ns = _load_namespace() + config = {"output": {"state_dir": "${HARNESS_DIR}/state/tech-hotspot-radar"}} + state_dir = ns["_browser_agent_state_dir"](config) + request_dir = ns["_browser_agent_request_dir"](config, "planner smoke") + expected_root = Path.home() / ".solar" / "harness" / "state" / "tech-hotspot-radar" + + assert state_dir == expected_root + assert request_dir.parent == expected_root / "browser-agent-requests" + assert "${HARNESS_DIR}" not in str(request_dir) + + +def test_call_browser_agent_chatgpt_text_reuses_ai_influence_scratch_conversation(monkeypatch, tmp_path): + wrapper = tmp_path / "fake_wrapper.py" + wrapper.write_text( + textwrap.dedent( + """ + import json + import os + import sys + from pathlib import Path + + req = Path(os.environ["BROWSER_AGENT_REQUEST_DIR"]) + req.mkdir(parents=True, exist_ok=True) + env_dump = { + "profile_id": os.environ.get("BROWSER_AGENT_PROFILE_ID"), + "chatgpt_url": os.environ.get("BROWSER_AGENT_CHATGPT_URL"), + "force_new_chat": os.environ.get("BROWSER_AGENT_CHATGPT_FORCE_NEW_CHAT"), + "require_isolated": os.environ.get("BROWSER_AGENT_CHATGPT_REQUIRE_ISOLATED_CONVERSATION"), + "open_project_first": os.environ.get("BROWSER_AGENT_CHATGPT_OPEN_PROJECT_FIRST"), + "require_project": os.environ.get("BROWSER_AGENT_CHATGPT_REQUIRE_PROJECT"), + } + (req / "fake-env.json").write_text(json.dumps(env_dump, ensure_ascii=False), encoding="utf-8") + target_url = os.environ.get("BROWSER_AGENT_CHATGPT_URL") or "https://chatgpt.com/c/scratch-shared-001" + (req / "page.json").write_text( + json.dumps({"url": target_url, "conversation_id": "scratch-conv-001"}, ensure_ascii=False), + encoding="utf-8", + ) + sys.stdout.write("x" * 1200) + """ + ), + encoding="utf-8", + ) + monkeypatch.setenv("TECH_HOTSPOT_BROWSER_CHATGPT_CMD", f"{sys.executable} {wrapper}") + ns = _load_namespace() + config = { + "output": { + "raw_dir": str(tmp_path / "raw"), + "state_dir": str(tmp_path / "state"), + }, + "youtube": { + "phase_report_reasoner": { + "headless": True, + "open_project_first": False, + "require_project": False, + "force_new_chat": False, + "require_isolated_conversation": False, + }, + "ai_influence_report_flow": { + "report_writer": { + "model": "chatgpt-5.5", + "headless": True, + "open_project_first": False, + "require_project": False, + "force_new_chat": False, + "require_isolated_conversation": False, + } + }, + }, + } + scratch_profile_id = ns["ai_influence_scratch_profile_id"](task_type="planner") + first = ns["call_browser_agent_chatgpt_text"]( + "first prompt", + config, + purpose="ai-influence-report-plan-2026-06-05", + expected="markdown", + scratch_profile_id=scratch_profile_id, + open_project_first=False, + require_project=False, + force_new_chat=False, + require_isolated_conversation=False, + ) + session_path = ns["_browser_agent_scratch_session_path"](config, scratch_profile_id) + session_payload = json.loads(session_path.read_text(encoding="utf-8")) + assert session_payload["conversation_url"] == "https://chatgpt.com/c/scratch-shared-001" + + second = ns["call_browser_agent_chatgpt_text"]( + "second prompt", + config, + purpose="ai-influence-report-plan-2026-06-05", + expected="markdown", + scratch_profile_id=scratch_profile_id, + open_project_first=False, + require_project=False, + force_new_chat=False, + require_isolated_conversation=False, + ) + env_dump = json.loads((Path(second["request_dir"]) / "fake-env.json").read_text(encoding="utf-8")) + assert env_dump["profile_id"] == scratch_profile_id + assert env_dump["chatgpt_url"] == "https://chatgpt.com/c/scratch-shared-001" + assert env_dump["force_new_chat"] == "false" + assert env_dump["require_isolated"] == "false" + assert env_dump["open_project_first"] == "false" + assert env_dump["require_project"] == "false" + + +def test_call_browser_agent_chatgpt_text_persists_scratch_session_before_timeout(monkeypatch, tmp_path): + wrapper = tmp_path / "slow_wrapper.py" + wrapper.write_text( + textwrap.dedent( + """ + import json + import os + import time + from pathlib import Path + + req = Path(os.environ["BROWSER_AGENT_REQUEST_DIR"]) + req.mkdir(parents=True, exist_ok=True) + (req / "post-submit-state.json").write_text( + json.dumps( + { + "url": "https://chatgpt.com/c/scratch-timeout-001", + "conversation_id": "scratch-timeout-001", + "is_generating": True, + }, + ensure_ascii=False, + ), + encoding="utf-8", + ) + time.sleep(5) + """ + ), + encoding="utf-8", + ) + monkeypatch.setenv("TECH_HOTSPOT_BROWSER_CHATGPT_CMD", f"{sys.executable} {wrapper}") + ns = _load_namespace() + config = { + "output": { + "raw_dir": str(tmp_path / "raw"), + "state_dir": str(tmp_path / "state"), + }, + "youtube": { + "phase_report_reasoner": { + "headless": True, + "open_project_first": False, + "require_project": False, + "force_new_chat": False, + "require_isolated_conversation": False, + }, + }, + } + scratch_profile_id = ns["ai_influence_scratch_profile_id"](task_type="planner") + with pytest.raises(subprocess.TimeoutExpired): + ns["call_browser_agent_chatgpt_text"]( + "slow prompt", + config, + purpose="ai-influence-timeout-persist-2026-06-05", + expected="json", + requested_timeout_seconds=1, + scratch_profile_id=scratch_profile_id, + open_project_first=False, + require_project=False, + force_new_chat=False, + require_isolated_conversation=False, + ) + + session_path = ns["_browser_agent_scratch_session_path"](config, scratch_profile_id) + session_payload = json.loads(session_path.read_text(encoding="utf-8")) + assert session_payload["conversation_url"] == "https://chatgpt.com/c/scratch-timeout-001" + assert session_payload["conversation_id"] == "scratch-timeout-001" + + +def test_call_browser_agent_chatgpt_text_skips_root_url_autopersist(monkeypatch, tmp_path): + wrapper = tmp_path / "root_wrapper.py" + wrapper.write_text( + textwrap.dedent( + """ + import json + import os + import time + from pathlib import Path + + req = Path(os.environ["BROWSER_AGENT_REQUEST_DIR"]) + req.mkdir(parents=True, exist_ok=True) + (req / "post-submit-state.json").write_text( + json.dumps( + { + "url": "https://chatgpt.com/", + "conversation_id": "", + "is_generating": True, + }, + ensure_ascii=False, + ), + encoding="utf-8", + ) + time.sleep(5) + """ + ), + encoding="utf-8", + ) + monkeypatch.setenv("TECH_HOTSPOT_BROWSER_CHATGPT_CMD", f"{sys.executable} {wrapper}") + ns = _load_namespace() + config = { + "output": { + "raw_dir": str(tmp_path / "raw"), + "state_dir": str(tmp_path / "state"), + }, + "youtube": { + "phase_report_reasoner": { + "headless": True, + "open_project_first": False, + "require_project": False, + "force_new_chat": False, + "require_isolated_conversation": False, + }, + }, + } + scratch_profile_id = ns["ai_influence_scratch_profile_id"](task_type="planner") + with pytest.raises(subprocess.TimeoutExpired): + ns["call_browser_agent_chatgpt_text"]( + "slow prompt", + config, + purpose="ai-influence-timeout-root-skip-2026-06-05", + expected="json", + requested_timeout_seconds=1, + scratch_profile_id=scratch_profile_id, + open_project_first=False, + require_project=False, + force_new_chat=False, + require_isolated_conversation=False, + ) + + session_path = ns["_browser_agent_scratch_session_path"](config, scratch_profile_id) + assert not session_path.exists() + + def test_planned_report_ir_builds_per_chapter_contract(): ns = _load_namespace() evidence_pack = { @@ -1048,6 +1304,7 @@ def test_planned_report_ir_builds_per_chapter_contract(): def test_chapter_prompt_requires_chapter_writer_only(): ns = _load_namespace() report_ir = { + "report_id": "agent-platform", "title": "Agent 平台化报告", "global_scope": "分析 agent runtime", "reader_value": "帮助判断趋势", @@ -1060,6 +1317,7 @@ def test_chapter_prompt_requires_chapter_writer_only(): "material_video_refs": ["V001"], } evidence = { + "date": "2026-05-31", "videos": [ { "video_ref": "V001", @@ -1080,6 +1338,11 @@ def test_chapter_prompt_requires_chapter_writer_only(): assert "不写整份报告" in prompt assert "### 工具接口协议化" in prompt assert "chapter_evidence_pack" in prompt + assert "[SCRATCH_CHAT_RESET_CONTRACT]" in prompt + assert "task_type: chapter_writer" in prompt + assert "report_id: agent-platform" in prompt + assert "chapter_id: ch_01" in prompt + assert "[CHAPTER_SCOPE_GUARD]" in prompt def test_build_planned_report_evidence_pack_skips_missing_status_transcript(tmp_path):