tagword-codeagent 1.1.6__py3-none-any.whl

codeagent/core/__init__.py

@@ -0,0 +1,2 @@
+from .models import Session  # noqa: F401
+

codeagent/core/attachments.py

@@ -0,0 +1,342 @@
+"""Attachment subsystem — file storage, user message building, directory scan."""
+
+from __future__ import annotations
+
+import fnmatch
+import mimetypes
+import os
+import re
+from dataclasses import asdict, dataclass, field
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any, Optional
+
+from codeagent.core import env as ca_env
+
+ATTACHMENT_TAG_RE = re.compile(
+    r"\[attachment:([^\s\]]+)(?:\s+([^\]]+))?\]",
+    re.IGNORECASE,
+)
+IMAGE_DIR_TAG_RE = re.compile(
+    r"\[image_dir:([^\s\]]+)(?:\s+max=(\d+))?\]",
+    re.IGNORECASE,
+)
+
+_DEFAULT_IMAGE_GLOBS = ("*.png", "*.jpg", "*.jpeg", "*.webp", "*.gif", "*.bmp")
+
+
+@dataclass
+class AttachmentMeta:
+    id: str
+    kind: str
+    mime: str
+    filename: str
+    size_bytes: int
+    path: str = ""
+    extracted_text: Optional[str] = None
+
+    def to_dict(self) -> dict[str, Any]:
+        return {k: v for k, v in asdict(self).items() if v is not None}
+
+
+def _max_count() -> int:
+    return ca_env.pick_int(8, "CODEAGENT_ATTACHMENTS_MAX_COUNT")
+
+
+def _max_bytes() -> int:
+    return ca_env.pick_int(20 * 1024 * 1024, "CODEAGENT_ATTACHMENTS_MAX_BYTES")
+
+
+def _dir_max_files() -> int:
+    return ca_env.pick_int(32, "CODEAGENT_ATTACHMENTS_DIR_MAX_FILES")
+
+
+def _allowed_mime_prefixes() -> tuple[str, ...]:
+    raw = ca_env.pick_default(
+        "image/,video/,audio/,application/pdf,text/",
+        "CODEAGENT_ATTACHMENTS_ALLOWED_MIME",
+    )
+    return tuple(p.strip() for p in raw.split(",") if p.strip())
+
+
+def mime_allowed(mime: str, filename: str = "") -> bool:
+    m = (mime or "").lower().split(";")[0].strip()
+    if not m and filename:
+        guessed, _ = mimetypes.guess_type(filename)
+        m = (guessed or "").lower()
+    if m in ("image/svg+xml", "text/html"):
+        return False
+    for prefix in _allowed_mime_prefixes():
+        if prefix.endswith("/") and m.startswith(prefix):
+            return True
+        if m == prefix:
+            return True
+    return False
+
+
+def _max_bytes_for_kind(kind: str) -> int:
+    k = (kind or "").lower()
+    if k in ("video", "audio"):
+        return ca_env.pick_int(100 * 1024 * 1024, "CODEAGENT_MEDIA_MAX_BYTES")
+    return _max_bytes()
+
+
+def _kind_for_mime(mime: str, filename: str) -> str:
+    m = (mime or "").lower()
+    if m.startswith("image/"):
+        return "image"
+    if m.startswith("video/"):
+        return "video"
+    if m.startswith("audio/"):
+        return "audio"
+    if m == "application/pdf" or filename.lower().endswith(".pdf"):
+        return "document"
+    return "document"
+
+
+def extract_document_text(raw: bytes, mime: str, filename: str = "") -> str:
+    m = (mime or "").lower()
+    name = (filename or "").lower()
+    if m.startswith("text/") or name.endswith((".txt", ".md", ".markdown")):
+        return raw.decode("utf-8", errors="replace").strip()
+    if m == "application/pdf" or name.endswith(".pdf"):
+        try:
+            from pypdf import PdfReader
+            import io
+
+            reader = PdfReader(io.BytesIO(raw))
+            parts = []
+            for page in reader.pages:
+                t = page.extract_text() or ""
+                if t.strip():
+                    parts.append(t.strip())
+            text = "\n\n".join(parts).strip()
+            if not text:
+                return "[PDF 无文本层；请转为图片后使用 vision_analyze]"
+            return text
+        except ImportError:
+            return "[PDF 解析需要 pip install pypdf]"
+        except Exception as e:
+            return f"[PDF 提取失败: {e}]"
+    return ""
+
+
+def save_attachment(
+    *,
+    agent_id: str,
+    session_id: str,
+    raw_bytes: bytes,
+    filename: str,
+    mime: str = "",
+) -> AttachmentMeta:
+    kind_guess = _kind_for_mime(mime, filename)
+    limit = _max_bytes_for_kind(kind_guess)
+    if len(raw_bytes) > limit:
+        raise ValueError(f"attachment exceeds {limit} bytes")
+    if not mime_allowed(mime, filename):
+        raise ValueError(f"mime not allowed: {mime or filename}")
+    from seed.core.media_store import save_session_media
+
+    aid, path = save_session_media(
+        agent_id=agent_id,
+        session_id=session_id,
+        raw_bytes=raw_bytes,
+        filename=filename,
+        mime=mime,
+    )
+    kind = _kind_for_mime(mime, filename)
+    extracted = None
+    if kind == "document":
+        extracted = extract_document_text(raw_bytes, mime, filename) or None
+    return AttachmentMeta(
+        id=aid,
+        kind=kind,
+        mime=(mime or mimetypes.guess_type(filename)[0] or "application/octet-stream"),
+        filename=filename or aid,
+        size_bytes=len(raw_bytes),
+        path=str(path),
+        extracted_text=extracted,
+    )
+
+
+def resolve_attachment_path(agent_id: str, session_id: str, attachment_id: str) -> Optional[Path]:
+    from seed.core.media_store import resolve_session_media_path
+
+    p = resolve_session_media_path(agent_id, session_id, attachment_id)
+    return p
+
+
+def load_attachment_meta(agent_id: str, session_id: str, attachment_id: str) -> Optional[AttachmentMeta]:
+    p = resolve_attachment_path(agent_id, session_id, attachment_id)
+    if not p or not p.is_file():
+        return None
+    mime, _ = mimetypes.guess_type(str(p))
+    kind = _kind_for_mime(mime or "", p.name)
+    return AttachmentMeta(
+        id=attachment_id,
+        kind=kind,
+        mime=mime or "application/octet-stream",
+        filename=p.name.split("_", 1)[-1] if "_" in p.name else p.name,
+        size_bytes=p.stat().st_size,
+        path=str(p.resolve()),
+    )
+
+
+def build_user_message(
+    text: str,
+    attachments: list[AttachmentMeta],
+    *,
+    image_dir_tag: str = "",
+) -> dict[str, Any]:
+    lines: list[str] = []
+    t = (text or "").strip()
+    if t:
+        lines.append(t)
+    elif attachments:
+        lines.append("[附件]")
+    for a in attachments:
+        if a.kind == "document" and a.extracted_text:
+            lines.append(f"[文档: {a.filename}]\n{a.extracted_text}")
+        else:
+            lines.append(f"[attachment:{a.id} {a.filename}]")
+    if image_dir_tag.strip():
+        lines.append(image_dir_tag.strip())
+    msg: dict[str, Any] = {
+        "role": "user",
+        "content": "\n".join(lines),
+        "ts": datetime.now(timezone.utc).isoformat(),
+    }
+    if attachments:
+        msg["attachments"] = [a.to_dict() for a in attachments]
+    return msg
+
+
+def content_text_for_skills(content: Any) -> str:
+    if content is None:
+        return ""
+    if isinstance(content, str):
+        text = content
+    elif isinstance(content, list):
+        parts = []
+        for item in content:
+            if isinstance(item, dict) and item.get("type") == "text":
+                parts.append(str(item.get("text") or ""))
+        text = "".join(parts)
+    else:
+        text = str(content)
+    text = ATTACHMENT_TAG_RE.sub("", text)
+    text = IMAGE_DIR_TAG_RE.sub("", text)
+    return text.strip()
+
+
+def parse_image_dir_tags(text: str) -> list[tuple[str, int]]:
+    out: list[tuple[str, int]] = []
+    for m in IMAGE_DIR_TAG_RE.finditer(text or ""):
+        rel = m.group(1).strip()
+        try:
+            mx = int(m.group(2)) if m.group(2) else _dir_max_files()
+        except ValueError:
+            mx = _dir_max_files()
+        out.append((rel, min(mx, _dir_max_files())))
+    return out
+
+
+def scan_image_directory(
+    workspace_root: Path,
+    rel_path: str,
+    *,
+    pattern: Optional[str] = None,
+    max_files: Optional[int] = None,
+) -> tuple[list[Path], bool]:
+    """Return (image paths, truncated). Paths resolved under workspace_root."""
+    root = workspace_root.resolve()
+    rel = (rel_path or "").strip().replace("\\", "/").lstrip("/")
+    if ".." in rel.split("/"):
+        raise ValueError("path traversal not allowed")
+    target = (root / rel).resolve()
+    if not str(target).startswith(str(root)):
+        raise ValueError("path outside workspace")
+    if not target.is_dir():
+        raise ValueError(f"not a directory: {rel_path}")
+
+    globs = _parse_globs(pattern)
+    limit = max_files if max_files is not None else _dir_max_files()
+    found: list[Path] = []
+    truncated = False
+    for dirpath, _, filenames in os.walk(target):
+        for name in sorted(filenames):
+            if not _matches_globs(name, globs):
+                continue
+            p = Path(dirpath) / name
+            if not p.is_file():
+                continue
+            found.append(p.resolve())
+            if len(found) >= limit:
+                truncated = True
+                return found, truncated
+    return found, truncated
+
+
+def _parse_globs(pattern: Optional[str]) -> tuple[str, ...]:
+    if pattern and pattern.strip():
+        return tuple(g.strip() for g in pattern.split(",") if g.strip())
+    raw = ca_env.pick_default(
+        ",".join(_DEFAULT_IMAGE_GLOBS),
+        "CODEAGENT_IMAGE_DIR_ALLOWED_GLOB",
+    )
+    return tuple(g.strip() for g in raw.split(",") if g.strip())
+
+
+def _matches_globs(name: str, globs: tuple[str, ...]) -> bool:
+    low = name.lower()
+    for g in globs:
+        if fnmatch.fnmatch(low, g.lower()):
+            return True
+    return False
+
+
+def message_has_image_attachments(msg: dict[str, Any]) -> bool:
+    atts = msg.get("attachments")
+    if isinstance(atts, list):
+        for a in atts:
+            if isinstance(a, dict) and a.get("kind") == "image":
+                return True
+    if IMAGE_DIR_TAG_RE.search(str(msg.get("content") or "")):
+        return True
+    return False
+
+
+def message_has_video_attachments(msg: dict[str, Any]) -> bool:
+    atts = msg.get("attachments")
+    if isinstance(atts, list):
+        for a in atts:
+            if isinstance(a, dict) and a.get("kind") == "video":
+                return True
+    return False
+
+
+def message_has_audio_attachments(msg: dict[str, Any]) -> bool:
+    atts = msg.get("attachments")
+    if isinstance(atts, list):
+        for a in atts:
+            if isinstance(a, dict) and a.get("kind") == "audio":
+                return True
+    return False
+
+
+def collect_attachment_ids_from_message(msg: dict[str, Any]) -> list[str]:
+    ids: list[str] = []
+    atts = msg.get("attachments")
+    if isinstance(atts, list):
+        for a in atts:
+            if isinstance(a, dict) and a.get("id"):
+                ids.append(str(a["id"]))
+    for m in ATTACHMENT_TAG_RE.finditer(str(msg.get("content") or "")):
+        ids.append(m.group(1))
+    seen: set[str] = set()
+    out: list[str] = []
+    for i in ids:
+        if i not in seen:
+            seen.add(i)
+            out.append(i)
+    return out

codeagent/core/audio_models.py

@@ -0,0 +1,50 @@
+"""Audio transcription preset resolution for audio_transcribe tool."""
+
+from __future__ import annotations
+
+from typing import Any, Optional
+
+from codeagent.core import env as ca_env
+
+
+def list_audio_presets() -> list[dict[str, Any]]:
+    from seed.core.llm_presets import load_presets
+
+    return [p for p in load_presets() if preset_supports_audio(p)]
+
+
+def preset_supports_audio(preset: dict[str, Any] | None) -> bool:
+    if not isinstance(preset, dict):
+        return False
+    return preset.get("supports_audio") is True
+
+
+def preset_supports_audio_id(preset_id: str) -> bool:
+    from seed.core.llm_presets import load_presets
+
+    pid = (preset_id or "").strip()
+    for p in load_presets():
+        if str(p.get("id") or "").strip() == pid:
+            return preset_supports_audio(p)
+    return False
+
+
+def resolve_audio_preset(preset_id: Optional[str] = None) -> dict[str, Any]:
+    from seed.core.llm_presets import load_presets
+
+    pid = (preset_id or "").strip()
+    if not pid:
+        pid = ca_env.pick_default("", "CODEAGENT_AUDIO_PRESET_ID").strip()
+    if not pid:
+        presets = list_audio_presets()
+        if len(presets) == 1:
+            return dict(presets[0])
+        raise ValueError(
+            "audio preset required: set CODEAGENT_AUDIO_PRESET_ID or Web UI audio_llm_id"
+        )
+    for p in load_presets():
+        if str(p.get("id") or "").strip() == pid:
+            if preset_supports_audio(p):
+                return dict(p)
+            raise ValueError(f"preset {pid} does not support audio transcription")
+    raise ValueError(f"audio preset not found: {pid}")

codeagent/core/bootstrap.py

@@ -0,0 +1,178 @@
+"""Code Agent process bootstrap (product home → config → Seed bridge)."""
+
+from __future__ import annotations
+
+import logging
+import os
+from pathlib import Path
+from typing import Optional
+
+from codeagent.core.env import apply_default_product_home
+from codeagent.core.seed_bridge import bridge_codeagent_env_to_seed
+from seed.integrations.env_config import apply_seed_env_from_config
+
+logger = logging.getLogger(__name__)
+
+_CODEAGENT_BOOTSTRAP_MD = """\
+# CodeAgent 首次启动引导
+
+首次使用请按以下顺序完成初始化。
+
+---
+
+## 1. 环境变量
+
+复制模板并填写 LLM 信息：
+
+```bash
+cp config/env.example config/env
+```
+
+最低必填项（编辑 `config/env`）：
+
+| 变量 | 说明 |
+|------|------|
+| `SEED_LLM_BASEURL` | LLM API 地址（OpenAI 格式） |
+| `SEED_LLM_MODEL` | 模型名称 |
+| `SEED_LLM_API_KEY` | API 密钥 |
+
+系统环境变量优先级 > env 文件。
+
+---
+
+## 2. Agent 人格配置
+
+Agent 行为由 `agents/default/persona/` 下的 Markdown 文件定义：
+
+| 文件 | 作用 |
+|------|------|
+| `identity.md` | 角色定位 |
+| `soul.md` | 核心价值观与行为准则 |
+| `agent.md` | 工作流程与项目管理规范 |
+| `skills.md` | 核心工作流 |
+| `tools.md` | 工具选用规则 |
+| `user.md` | 用户偏好 |
+
+按需修改即可调整 Agent 风格。
+
+---
+
+## 3. 首次验证
+
+```bash
+# Web UI（推荐）
+codeagent web
+
+# 或 CLI 模式
+codeagent chat
+```
+
+确认 LLM 能正常响应、会话日志写入正常。
+
+---
+
+## 4. （可选）Git 版本管理
+
+对配置进行版本控制，方便回溯：
+
+```bash
+cd ~/.codeagent
+git init
+```
+
+创建 `.gitignore` 忽略运行数据：
+
+```
+agents/default/sessions/
+agents/default/projects-data/
+agents/default/memory/
+mcp-minimax-out/
+config/env
+config/seed.env
+config/webui.token
+config/codeagent.webui.token
+```
+
+首次提交：
+
+```bash
+git add .gitignore agents/default/persona/ agents/default/skills/ config/
+git commit -m "chore: init codeagent config"
+```
+
+---
+
+## 目录结构
+
+```
+~/.codeagent/
+├── config/                  # 全局配置（env、cron、MCP、LSP、hooks）
+├── agents/default/
+│   ├── persona/             # Agent 人格定义（修改这里）
+│   ├── skills/              # 自定义技能
+│   ├── sessions/            # 对话历史（自动生成）
+│   ├── memory/              # 长期记忆（自动生成）
+│   └── projects-data/       # 项目记忆（自动生成）
+├── mcp-minimax-out/         # MCP 工具缓存
+└── bootstrap.md             # ← 本文件
+```
+"""
+
+
+def _ensure_codeagent_bootstrap_md(home: Path) -> None:
+    """Overwrite Seed's generic bootstrap.md with CodeAgent's own version."""
+    dest = home / "config" / "bootstrap.md"
+    dest.parent.mkdir(parents=True, exist_ok=True)
+    # Always overwrite: CodeAgent's bootstrap is the authoritative one.
+    dest.write_text(_CODEAGENT_BOOTSTRAP_MD, encoding="utf-8")
+
+
+def bootstrap_codeagent_runtime(base: Optional[Path] = None) -> Path:
+    """
+    1. Default data root ``~/.codeagent`` (or ``CODEAGENT_HOME``)
+    2. Set ``SEED_PROJECT_ROOT`` so Seed / seed-tools use the same tree
+    3. Load ``config/env`` (or legacy ``config/seed.env``) + ``config/codeagent.env``
+    4. Bridge ``CODEAGENT_*`` → ``SEED_*`` for kernel env keys
+    5. Write CodeAgent's bootstrap.md (overrides Seed's generic version)
+    """
+    if base is not None:
+        home = base.resolve()
+        os.environ["CODEAGENT_HOME"] = str(home)
+        os.environ["SEED_PROJECT_ROOT"] = str(home)
+        home.mkdir(parents=True, exist_ok=True)
+        (home / "config").mkdir(parents=True, exist_ok=True)
+    else:
+        home = apply_default_product_home()
+
+    apply_seed_env_from_config(None)
+    bridge_codeagent_env_to_seed()
+
+    from codeagent.runtime.compact_prompt import default_summarizer_prompt_path
+
+    if default_summarizer_prompt_path().is_file():
+        os.environ.setdefault(
+            "CODEAGENT_CONTEXT_COMPACT_SUMMARIZER_PROMPT_FILE",
+            str(default_summarizer_prompt_path()),
+        )
+    os.environ.setdefault("CODEAGENT_CONTEXT_COMPACT_MID_LOOP", "1")
+    os.environ.setdefault("CODEAGENT_CONTEXT_COMPACT_KEEP_USER_ROUNDS", "2")
+    bridge_codeagent_env_to_seed()
+
+    _ensure_codeagent_bootstrap_md(home)
+
+    # ── Team config: load if exists ──
+    try:
+        team_cfg_path = home / "config" / "team.json"
+        if team_cfg_path.is_file():
+            from codeagent.core.team_manager import TeamManager
+            tm = TeamManager()
+            if tm.init_from_config(team_cfg_path):
+                logger.info(f"Team mode enabled: {len(tm.config.members)} member agents configured")
+            else:
+                logger.info("Single-agent mode (team config invalid or empty)")
+        else:
+            logger.info("Single-agent mode (no team.json found)")
+    except Exception:
+        logger.exception("Team config init failed — continuing in single-agent mode")
+
+    return home