tools/notion_tool.py

"""
    backend/tools/notion_tool.py — P24-F2: Notion read/write/search/append.

    Permette all'agente di leggere, scrivere e cercare in Notion.
    Richiede NOTION_TOKEN env var (Integration Token da notion.so/my-integrations).

    Operazioni:
      search  — cerca pagine/database per testo
      read    — legge contenuto di una pagina (restituisce markdown-like)
      write   — crea nuova pagina sotto un genitore
      append  — aggiunge testo/markdown a pagina esistente
    """
from __future__ import annotations

import logging
import os
from typing import Any

import httpx

_logger = logging.getLogger("agente_ai.tools.notion_tool")

# Token resolution: env var → vault (fallback) → empty
def _resolve_notion_token() -> str:
    """Legge NOTION_TOKEN da env var o dal vault crittografato del backend."""
    # 1. Env var diretta (Railway secrets, HF Spaces secrets, .env locale)
    token = os.getenv("NOTION_TOKEN", "")
    if token:
        return token
    # 2. Fallback: vault crittografato (persistente tra i restart)
    try:
        import json as _j
        from pathlib import Path as _P
        import base64 as _b64, hashlib as _h, hmac as _hmac_m
        vault_path_str = os.getenv("VAULT_PATH", "/tmp/.vault_data/vault_secrets.json")
        _vp = _P(vault_path_str)
        if _vp.exists():
            data = _j.loads(_vp.read_text())
            raw_enc = data.get("notion_token", "")
            if raw_enc:
                raw_key = os.getenv("VAULT_KEY", "").encode() or b""
                if raw_key:
                    _vk = _h.sha256(raw_key).digest()
                    raw = _b64.b64decode(raw_enc)
                    sig, ct = raw[:32], raw[32:]
                    if _hmac_m.compare_digest(sig, _hmac_m.new(_vk, ct, _h.sha256).digest()):
                        ks = b""
                        for i in range((len(ct) // 32) + 1):
                            ks += _h.sha256(_vk + i.to_bytes(4, "big")).digest()
                        return bytes(a ^ b for a, b in zip(ct, ks[:len(ct)])).decode("utf-8")
    except Exception:  # noqa: BLE001
        pass
    return ""

_BASE_URL     = "https://api.notion.com/v1"
_NOTION_VER   = "2022-06-28"
_TIMEOUT      = 15.0
_MAX_TEXT     = 8000   # max caratteri estratti da una pagina


def _hdrs() -> dict[str, str]:
    # Risoluzione dinamica del token per riflettere aggiornamenti del Vault senza restart
    token = _resolve_notion_token()
    return {
        "Authorization":  f"Bearer {token}",
        "Notion-Version": _NOTION_VER,
        "Content-Type":   "application/json",
    }


# ── Parsing blocchi → testo ────────────────────────────────────────────────────

def _blocks_to_text(blocks: list[dict]) -> str:
    lines: list[str] = []
    for b in blocks:
        btype = b.get("type", "")
        obj   = b.get(btype, {})
        rich  = obj.get("rich_text", [])
        text  = "".join(r.get("plain_text", "") for r in rich)
        if btype == "code":
            lang = obj.get("language", "")
            lines.append(f"```{lang}\n{text}\n```")
        elif btype == "heading_1":
            lines.append(f"# {text}")
        elif btype == "heading_2":
            lines.append(f"## {text}")
        elif btype == "heading_3":
            lines.append(f"### {text}")
        elif btype == "bulleted_list_item":
            lines.append(f"• {text}")
        elif btype == "numbered_list_item":
            lines.append(f"- {text}")
        elif btype == "to_do":
            mark = "✅" if obj.get("checked") else "☐"
            lines.append(f"{mark} {text}")
        elif btype == "divider":
            lines.append("---")
        elif text:
            lines.append(text)
    return "\n".join(lines)


# ── Markdown → blocchi Notion ──────────────────────────────────────────────────

def _text_to_blocks(text: str) -> list[dict[str, Any]]:
    blocks: list[dict[str, Any]] = []
    in_code   = False
    code_buf: list[str] = []
    code_lang = "plain text"

    def _rt(s: str) -> list[dict]:
        return [{"type": "text", "text": {"content": s[:2000]}}]

    for line in text.split("\n"):
        if line.startswith("```"):
            if not in_code:
                in_code = True
                code_lang = line[3:].strip() or "plain text"
                code_buf  = []
            else:
                in_code = False
                blocks.append({"type": "code", "code": {
                    "language": code_lang,
                    "rich_text": _rt("\n".join(code_buf)),
                }})
            continue
        if in_code:
            code_buf.append(line)
            continue

        if line.startswith("# "):
            blocks.append({"type": "heading_1", "heading_1": {"rich_text": _rt(line[2:])}})
        elif line.startswith("## "):
            blocks.append({"type": "heading_2", "heading_2": {"rich_text": _rt(line[3:])}})
        elif line.startswith("### "):
            blocks.append({"type": "heading_3", "heading_3": {"rich_text": _rt(line[4:])}})
        elif line.startswith(("- ", "• ")):
            blocks.append({"type": "bulleted_list_item", "bulleted_list_item": {"rich_text": _rt(line[2:])}})
        elif line.strip() in ("---", "***", "___"):
            blocks.append({"type": "divider", "divider": {}})
        elif not line.strip():
            blocks.append({"type": "paragraph", "paragraph": {"rich_text": []}})
        else:
            blocks.append({"type": "paragraph", "paragraph": {"rich_text": _rt(line)}})

    return blocks[:100]   # Notion: max 100 blocchi per richiesta


# ── Funzione principale ────────────────────────────────────────────────────────

async def notion_rw(
    action: str,
    query: str          = "",
    page_id: str        = "",
    parent_id: str      = "",
    title: str          = "",
    content: str        = "",
    max_results: int    = 5,
) -> dict[str, Any]:
    """
    Interagisce con Notion (leggi / scrivi / cerca / appendi).

    action = "search"  → cerca pagine per testo (query obbligatorio)
    action = "read"    → legge contenuto pagina (page_id obbligatorio)
    action = "write"   → crea nuova pagina (parent_id + title; content opzionale)
    action = "append"  → aggiunge contenuto a pagina esistente (page_id + content)

    Richiede NOTION_TOKEN nel vault del backend o nelle env vars.
    """
    # Verifica token all'inizio della chiamata
    if not _resolve_notion_token():
        return {
            "ok":    False,
            "error": (
                "NOTION_TOKEN non configurato. "
                "Aggiungilo come NOTION_TOKEN nei secrets Railway/HF Spaces, "
                "oppure salvalo nel vault tramite /api/vault (chiave: notion_token)."
            ),
        }

    action = action.lower().strip()

    try:
        async with httpx.AsyncClient(timeout=_TIMEOUT, base_url=_BASE_URL) as c:

            # ── SEARCH ────────────────────────────────────────────────────────
            if action == "search":
                if not query:
                    return {"ok": False, "error": "query è obbligatorio per action=search"}
                r = await c.post("/search", headers=_hdrs(),
                    json={"query": query, "page_size": min(max(1, max_results), 20)})
                if r.status_code != 200:
                    return {"ok": False, "error": f"Notion {r.status_code}: {r.text[:300]}"}
                items = []
                for obj in r.json().get("results", []):
                    otype = obj.get("object", "")
                    t = ""
                    if otype == "page":
                        for v in obj.get("properties", {}).values():
                            if v.get("type") == "title":
                                t = "".join(x.get("plain_text","") for x in v.get("title",[]))
                                break
                    elif otype == "database":
                        t = "".join(x.get("plain_text","") for x in obj.get("title",[]))
                    items.append({
                        "id":          obj.get("id","").replace("-",""),
                        "title":       t or "(senza titolo)",
                        "type":        otype,
                        "url":         obj.get("url",""),
                        "last_edited": obj.get("last_edited_time",""),
                    })
                return {"ok": True, "action": "search", "query": query,
                        "results": items, "count": len(items)}

            # ── READ ──────────────────────────────────────────────────────────
            elif action == "read":
                if not page_id:
                    return {"ok": False, "error": "page_id è obbligatorio per action=read"}
                pid = page_id.replace("-","")
                meta = await c.get(f"/pages/{pid}", headers=_hdrs())
                if meta.status_code != 200:
                    return {"ok": False, "error": f"Pagina non trovata o accesso negato ({meta.status_code})"}
                m = meta.json()
                t = ""
                for v in m.get("properties", {}).values():
                    if v.get("type") == "title":
                        t = "".join(x.get("plain_text","") for x in v.get("title",[]))
                        break
                blk_r = await c.get(f"/blocks/{pid}/children",
                    headers=_hdrs(), params={"page_size": 100})
                blocks = blk_r.json().get("results",[]) if blk_r.status_code == 200 else []
                return {
                    "ok": True, "action": "read",
                    "page_id": pid, "title": t,
                    "url": m.get("url",""),
                    "last_edited": m.get("last_edited_time",""),
                    "content": _blocks_to_text(blocks)[:_MAX_TEXT],
                    "block_count": len(blocks),
                }

            # ── WRITE ─────────────────────────────────────────────────────────
            elif action == "write":
                if not parent_id:
                    return {"ok": False, "error": "parent_id obbligatorio per action=write"}
                if not title:
                    return {"ok": False, "error": "title obbligatorio per action=write"}
                pid    = parent_id.replace("-","")
                blocks = _text_to_blocks(content) if content else []
                payload: dict[str, Any] = {
                    "parent":     {"page_id": pid},
                    "properties": {"title": {"title": [{"type":"text","text":{"content":title[:2000]}}]}},
                    "children":   blocks,
                }
                r = await c.post("/pages", headers=_hdrs(), json=payload)
                if r.status_code not in (200, 201):
                    return {"ok": False, "error": f"Notion {r.status_code}: {r.text[:300]}"}
                created = r.json()
                return {
                    "ok": True, "action": "write",
                    "page_id": created.get("id","").replace("-",""),
                    "url":     created.get("url",""),
                    "title":   title,
                    "blocks_written": len(blocks),
                    "message": f"Pagina '{title}' creata su Notion",
                }

            # ── APPEND ────────────────────────────────────────────────────────
            elif action == "append":
                if not page_id:
                    return {"ok": False, "error": "page_id obbligatorio per action=append"}
                if not content:
                    return {"ok": False, "error": "content obbligatorio per action=append"}
                pid    = page_id.replace("-","")
                blocks = _text_to_blocks(content)
                r = await c.patch(f"/blocks/{pid}/children",
                    headers=_hdrs(), json={"children": blocks})
                if r.status_code != 200:
                    return {"ok": False, "error": f"Notion {r.status_code}: {r.text[:300]}"}
                return {
                    "ok": True, "action": "append",
                    "page_id": pid,
                    "blocks_appended": len(blocks),
                    "message": f"{len(blocks)} blocchi aggiunti alla pagina",
                }

            else:
                return {"ok": False, "error": f"action non valida: '{action}'. Valori: search, read, write, append"}

    except httpx.TimeoutException:
        return {"ok": False, "error": f"Timeout ({_TIMEOUT:.0f}s) — Notion API non ha risposto"}
    except Exception as exc:  # noqa: BLE001
        _logger.warning("[notion_tool] %s", exc)
        return {"ok": False, "error": str(exc)[:300]}


# ── Tool registry descriptor ───────────────────────────────────────────────────
TOOL_DESCRIPTOR = {
    "name": "notion_rw",
    "description": (
        "Leggi, scrivi e cerca su Notion. "
        "action=search: cerca pagine per query testuale. "
        "action=read: legge contenuto pagina (page_id). "
        "action=write: crea nuova pagina (parent_id + title + content opzionale). "
        "action=append: aggiunge testo/markdown a pagina esistente (page_id + content). "
        "Richiede NOTION_TOKEN configurato nel backend."
    ),
    "parameters": {
        "type": "object",
        "properties": {
            "action":      {"type": "string", "enum": ["search","read","write","append"],
                            "description": "Operazione da eseguire"},
            "query":       {"type": "string",  "description": "Testo di ricerca (per search)"},
            "page_id":     {"type": "string",  "description": "ID pagina Notion (per read/append)"},
            "parent_id":   {"type": "string",  "description": "ID pagina genitore (per write)"},
            "title":       {"type": "string",  "description": "Titolo nuova pagina (per write)"},
            "content":     {"type": "string",  "description": "Testo/Markdown da scrivere o appendere"},
            "max_results": {"type": "integer", "description": "Max risultati search (default 5, max 20)"},
        },
        "required": ["action"],
    },
    "fn": notion_rw,
}