初始化：独立 oclaw 仓库首提交

- 在 oclaw/ 下重新初始化 Git 仓库
- 补齐子仓库 .gitignore，避免提交本地运行态数据（_local、node_modules、logs 等）
- 提交当前工程代码与配置

Made-with: Cursor

prompts/README.md

@@ -0,0 +1,22 @@
+---
+title: prompts_scope_after_openclaw_migration
+summary: Scope boundary for legacy and fallback prompts under oclaw/prompts.
+read_when:
+  - Maintaining prompt files under oclaw/prompts.
+  - Distinguishing OpenClaw runtime prompts from non-OpenClaw prompts.
+---
+
+# `oclaw/prompts` Scope
+
+OpenClaw runtime prompts have been migrated to `oclaw/prompts_openclaw`.
+
+`oclaw/prompts` now keeps only:
+
+- `fallback/*`: user-facing fallback/error templates.
+- Non-OpenClaw shared templates still used by runtime components, such as:
+  - `runtime/default_system.{zh,en}.md`
+  - `tools/*`
+  - `image/default_edit_prompt.zh.md`
+
+Do not add new OpenClaw manager/router/specialist/runtime chain prompts here.
+Use `oclaw/prompts_openclaw/*` instead.

prompts/__init__.py

@@ -0,0 +1,3 @@
+from .loader import render_openclaw_prompt, render_prompt, render_prompt_for_lang
+
+__all__ = ["render_prompt", "render_prompt_for_lang", "render_openclaw_prompt"]

prompts/fallback/ollama_failure.en.md

@@ -0,0 +1,9 @@
+---
+title: fallback_ollama_failure_en
+summary: ollama transport failure banner in English
+read_when: ollama transport exception
+---
+**Could not reach the Ollama / compatible endpoint:** `{{error_type}}: {{error_message}}`
+
+_Falling back to the built-in rule engine for this reply._
+

prompts/fallback/ollama_failure.zh.md

@@ -0,0 +1,9 @@
+---
+title: fallback_ollama_failure_zh
+summary: ollama transport failure banner in Chinese
+read_when: ollama transport exception
+---
+**无法连接本地 Ollama（或兼容接口）：** `{{error_type}}: {{error_message}}`
+
+_本次回答已改用内置规则引擎兜底。_
+

prompts/fallback/openai_missing_key_user.en.md

@@ -0,0 +1,11 @@
+---
+title: fallback_openai_missing_key_user_en
+summary: user-facing message when openai key is missing
+read_when: openai mode selected but no api key
+---
+**Cannot use this profile:** no API Key is set for the OpenAI-compatible endpoint, and `OPENAI_API_KEY` is not in the environment.
+
+Add a key in **Settings** and save, or switch to the built-in **Local Ollama (default)** profile.
+
+_No request was sent to the model service._
+

prompts/fallback/openai_missing_key_user.zh.md

@@ -0,0 +1,11 @@
+---
+title: fallback_openai_missing_key_user_zh
+summary: user-facing message when openai key is missing
+read_when: openai mode selected but no api key
+---
+**当前配置无法接入模型：** 本配置为 OpenAI / 兼容 API，但未填写 API Key，且环境中也没有 `OPENAI_API_KEY`。
+
+请在「设置」中填写 Key 并保存，或切换到内置的「本地 Ollama（默认）」配置后再试。
+
+_本次未向模型服务发送请求。_
+

prompts/fallback/openai_transport_error.en.md

@@ -0,0 +1,9 @@
+---
+title: fallback_openai_transport_error_en
+summary: generic OpenAI-compatible transport error in English
+read_when: model request failed
+---
+**Model request failed:** `{{error_type}}: {{error_message}}`
+
+Check API key, base URL, and network, then try again.{{extra_tail}}
+

prompts/fallback/openai_transport_error.zh.md

@@ -0,0 +1,9 @@
+---
+title: fallback_openai_transport_error_zh
+summary: generic OpenAI-compatible transport error in Chinese
+read_when: model request failed
+---
+**调用模型失败：** `{{error_type}}: {{error_message}}`
+
+请检查 API Key、Base URL 与网络后重试。{{extra_tail}}
+

prompts/fallback/openai_transport_oversized.en.md

@@ -0,0 +1,9 @@
+---
+title: fallback_openai_transport_oversized_en
+summary: oversized payload failure in English
+read_when: model request too large
+---
+**Model request failed (message too large):** `{{error_type}}: {{error_message}}`
+
+Common causes: oversized tools schema or large tool results in history. Tune `AIA_OPENAI_TOOLS_MAX_JSON_CHARS`, reduce MCP tools, or narrow the query.
+

prompts/fallback/openai_transport_oversized.zh.md

@@ -0,0 +1,9 @@
+---
+title: fallback_openai_transport_oversized_zh
+summary: oversized payload failure in Chinese
+read_when: model request too large
+---
+**调用模型失败（单条消息过长）：** `{{error_type}}: {{error_message}}`
+
+常见原因：tools 负载过大或 tool 返回过长。可调整 `AIA_OPENAI_TOOLS_MAX_JSON_CHARS`，或减少 MCP / 缩小查询范围后重试。
+

prompts/fallback/openclaw_runtime_error.en.md

@@ -0,0 +1,7 @@
+---
+title: fallback_openclaw_runtime_error_en
+summary: fallback text when openclaw runtime failed
+read_when: gateway exception path
+---
+oclaw runtime error: executor missing model/tools
+

prompts/fallback/openclaw_runtime_error.zh.md

@@ -0,0 +1,7 @@
+---
+title: fallback_openclaw_runtime_error_zh
+summary: fallback text when openclaw runtime failed
+read_when: gateway exception path
+---
+oclaw 运行失败：执行器缺少 model/tools
+

prompts/fallback/task_queued.en.md

@@ -0,0 +1,7 @@
+---
+title: fallback_task_queued_en
+summary: User-visible text when async task is enqueued.
+read_when: gateway async_task mode
+---
+
+Task queued, task_id={{task_id}}

prompts/fallback/task_queued.zh.md

@@ -0,0 +1,7 @@
+---
+title: fallback_task_queued_zh
+summary: User-visible text when async task is enqueued.
+read_when: gateway async_task mode
+---
+
+任务已入队，task_id={{task_id}}

prompts/frontmatter.py

@@ -0,0 +1,92 @@
+from __future__ import annotations
+
+import json
+import os
+from typing import Any
+
+try:
+    import yaml
+except ImportError:  # pragma: no cover - requirements.txt includes PyYAML
+    yaml = None  # type: ignore[assignment]
+
+
+def split_markdown_frontmatter(raw: str) -> tuple[str, str]:
+    """Split leading YAML frontmatter from markdown body (oclaw-style `---` fences)."""
+    txt = str(raw or "")
+    if not txt.startswith("---\n"):
+        return "", txt
+    end = txt.find("\n---\n", 4)
+    if end < 0:
+        return "", txt
+    return txt[4:end], txt[end + 5 :]
+
+
+def _legacy_line_parse(frontmatter_text: str) -> dict[str, Any]:
+    """Best-effort key: value lines (pre-YAML migration compatibility)."""
+    out: dict[str, Any] = {}
+    for ln in str(frontmatter_text or "").splitlines():
+        line = ln.strip()
+        if not line or line.startswith("#") or ":" not in line:
+            continue
+        k, v = line.split(":", 1)
+        key = str(k).strip()
+        val = str(v).strip()
+        if (val.startswith('"') and val.endswith('"')) or (val.startswith("'") and val.endswith("'")):
+            val = val[1:-1]
+        out[key] = val
+    meta_raw = str(out.get("metadata") or "").strip()
+    if meta_raw:
+        try:
+            out["metadata"] = json.loads(meta_raw)
+        except Exception:
+            out["metadata"] = {}
+    else:
+        out["metadata"] = {}
+    return out
+
+
+def frontmatter_strict_yaml() -> bool:
+    return str(os.getenv("AIA_PROMPT_FRONTMATTER_STRICT", "")).strip().lower() in {"1", "true", "yes", "on"}
+
+
+def parse_frontmatter_dict(frontmatter_text: str, *, source: str = "prompt") -> dict[str, Any]:
+    """
+    Parse YAML frontmatter. On YAML failure, falls back to legacy line parser unless
+    AIA_PROMPT_FRONTMATTER_STRICT is truthy (then raises).
+    """
+    text = str(frontmatter_text or "").strip()
+    if not text:
+        return {}
+    strict = frontmatter_strict_yaml()
+    if yaml is None:
+        if strict:
+            raise RuntimeError("PyYAML is required when AIA_PROMPT_FRONTMATTER_STRICT is enabled")
+        return _legacy_line_parse(text)
+    try:
+        data = yaml.safe_load(text)
+    except yaml.YAMLError as exc:
+        if strict:
+            raise ValueError(f"invalid YAML frontmatter ({source}): {exc}") from exc
+        return _legacy_line_parse(text)
+    if data is None:
+        return {}
+    if not isinstance(data, dict):
+        if strict:
+            raise ValueError(f"frontmatter YAML root must be a mapping ({source})")
+        return _legacy_line_parse(text)
+    return dict(data)
+
+
+def parse_markdown_document(raw: str, *, source: str = "markdown") -> tuple[dict[str, Any], str]:
+    """Split file and parse frontmatter to a dict; body is stripped markdown."""
+    fm_text, body = split_markdown_frontmatter(raw)
+    meta = parse_frontmatter_dict(fm_text, source=source) if fm_text.strip() else {}
+    return meta, str(body or "").strip()
+
+
+__all__ = [
+    "frontmatter_strict_yaml",
+    "parse_frontmatter_dict",
+    "parse_markdown_document",
+    "split_markdown_frontmatter",
+]

prompts/image/default_edit_prompt.zh.md

@@ -0,0 +1,7 @@
+---
+title: image_default_edit_prompt_zh
+summary: Default image editing prompt text.
+read_when: image prompt is empty
+---
+请基于输入图片和用户要求完成图像编辑，并返回结果图片。
+

prompts/loader.py

@@ -0,0 +1,97 @@
+from __future__ import annotations
+
+import re
+from dataclasses import dataclass
+from functools import lru_cache
+from pathlib import Path
+from typing import Any
+
+from oclaw.platform.config.paths import PROJECT_ROOT
+from oclaw.prompts.frontmatter import parse_markdown_document
+
+_VAR_RE = re.compile(r"\{\{\s*([a-zA-Z0-9_]+)\s*\}\}")
+
+
+@dataclass(frozen=True)
+class PromptDoc:
+    frontmatter: dict[str, Any]
+    body: str
+
+
+def _prompts_root() -> Path:
+    return (PROJECT_ROOT / "oclaw" / "prompts").resolve()
+
+
+def _openclaw_prompts_root() -> Path:
+    return (PROJECT_ROOT / "oclaw" / "prompts_openclaw").resolve()
+
+
+@lru_cache(maxsize=512)
+def load_prompt_doc(prompt_path: str) -> PromptDoc:
+    p = (_prompts_root() / str(prompt_path)).resolve()
+    if not p.exists():
+        raise FileNotFoundError(f"prompt not found: {prompt_path}")
+    raw = p.read_text(encoding="utf-8")
+    fm, body = parse_markdown_document(raw, source=f"prompt:{prompt_path}")
+    return PromptDoc(frontmatter=fm, body=body)
+
+
+@lru_cache(maxsize=512)
+def load_openclaw_prompt_doc(prompt_path: str) -> PromptDoc:
+    p = (_openclaw_prompts_root() / str(prompt_path)).resolve()
+    if not p.exists():
+        raise FileNotFoundError(f"openclaw prompt not found: {prompt_path}")
+    raw = p.read_text(encoding="utf-8")
+    fm, body = parse_markdown_document(raw, source=f"openclaw_prompt:{prompt_path}")
+    return PromptDoc(frontmatter=fm, body=body)
+
+
+def render_prompt_for_lang(stem: str, lang: str, *, variables: dict[str, Any] | None = None, strict: bool = True) -> str:
+    suf = "en" if (lang or "zh").strip().lower().startswith("en") else "zh"
+    return render_prompt(f"{stem}.{suf}.md", variables=variables, strict=strict)
+
+
+def render_prompt(prompt_path: str, *, variables: dict[str, Any] | None = None, strict: bool = True) -> str:
+    doc = load_prompt_doc(prompt_path)
+    vars_in = dict(variables or {})
+    missing: set[str] = set()
+
+    def _repl(m: re.Match[str]) -> str:
+        key = str(m.group(1) or "")
+        if key in vars_in:
+            return str(vars_in.get(key) or "")
+        missing.add(key)
+        return ""
+
+    out = _VAR_RE.sub(_repl, doc.body)
+    if strict and missing:
+        raise ValueError(f"missing prompt variables for {prompt_path}: {', '.join(sorted(missing))}")
+    return out.strip()
+
+
+def render_openclaw_prompt(prompt_path: str, *, variables: dict[str, Any] | None = None, strict: bool = True) -> str:
+    doc = load_openclaw_prompt_doc(prompt_path)
+    vars_in = dict(variables or {})
+    missing: set[str] = set()
+
+    def _repl(m: re.Match[str]) -> str:
+        key = str(m.group(1) or "")
+        if key in vars_in:
+            return str(vars_in.get(key) or "")
+        missing.add(key)
+        return ""
+
+    out = _VAR_RE.sub(_repl, doc.body)
+    if strict and missing:
+        raise ValueError(f"missing openclaw prompt variables for {prompt_path}: {', '.join(sorted(missing))}")
+    return out.strip()
+
+
+__all__ = [
+    "PromptDoc",
+    "load_prompt_doc",
+    "load_openclaw_prompt_doc",
+    "render_prompt",
+    "render_prompt_for_lang",
+    "render_openclaw_prompt",
+]

prompts/runtime/default_system.en.md

@@ -0,0 +1,27 @@
+---
+title: default_runtime_system_en
+summary: Default system prompt for general runtime (English template).
+read_when: agent runtime start
+---
+
+# Identity
+
+You are a general-purpose AI assistant.
+
+## Input Constraints
+
+- Answer in the same language as the user.
+
+## Execution Rules
+
+- When external data or actions are needed, prefer the tools provided by the system.
+- Tool calls are carried by the platform protocol; do not paste tool-protocol JSON in the user-visible reply.
+
+## Output Format
+
+- Produce user-readable text; conclusions first, steps concise.
+
+## Safety
+
+- Do not fabricate tool results.
+- When uncertain, state limits and ask for missing information.

prompts/runtime/default_system.zh.md

@@ -0,0 +1,22 @@
+---
+title: default_runtime_system_zh
+summary: Default system prompt for general runtime.
+read_when: agent runtime start
+---
+# Identity
+你是一个通用 AI 助手。
+
+## Input Constraints
+- 使用用户输入的语言回答。
+
+## Execution Rules
+- 当需要外部数据或执行动作时，优先调用系统提供的工具。
+- 工具调用由平台协议承载，不要在正文里输出工具协议 JSON。
+
+## Output Format
+- 直接输出用户可读内容，结论优先，步骤简洁。
+
+## Safety
+- 不要伪造工具结果。
+- 不确定时先说明边界并请求补充信息。
+

prompts/tools/image_attachment_meta.md

@@ -0,0 +1,8 @@
+---
+title: image_attachment_meta_block
+summary: Metadata-only image attachment block for LLM message context.
+read_when: building user message content
+---
+--- Image Attachment (meta only) ---
+{{meta_line}}
+

prompts/tools/text_attachment_wrap.md

@@ -0,0 +1,9 @@
+---
+title: text_attachment_wrap_block
+summary: Wrapper for text file attachments in LLM context.
+read_when: building user message content
+---
+--- File: {{name}} ---
+{{content}}
+--- End of File ---
+

prompts/tools/tool_result_unpaired.md

@@ -0,0 +1,8 @@
+---
+title: tool_result_unpaired_block
+summary: Downgraded tool result block when call id pairing is broken.
+read_when: building assistant fallback message
+---
+[{{tag}}]
+{{payload}}
+