Agentic-OS/backend/app/services/investigation_context.py

"""Load prior knowledge from project-memory and Obsidian before agent triage."""
from __future__ import annotations

import asyncio
import json
import logging
import os
import re
from typing import Any

from app.config import settings
from app.services import integrations

logger = logging.getLogger(__name__)

_STOPWORDS = frozenset(
    """
    a an the and or but in on at to for of is are was were be been being
    with from by as it this that these those i we you they he she
    check status health task issue vessel ship
    """.split()
)

_MAX_MEMORY_HITS = 8
_MAX_OBSIDIAN_HITS = 6
_EXCERPT_CHARS = 400


def extract_search_keywords(text: str, *, max_terms: int = 8) -> list[str]:
    """Pull meaningful terms from title + issue for memory/Obsidian search."""
    raw = re.findall(r"[a-zA-Z0-9][a-zA-Z0-9._/-]{1,}", text.lower())
    terms: list[str] = []
    seen: set[str] = set()
    for tok in raw:
        if tok in _STOPWORDS or len(tok) < 3:
            continue
        if tok.isdigit():
            continue
        if tok not in seen:
            seen.add(tok)
            terms.append(tok)
        if len(terms) >= max_terms:
            break
    return terms


def _parse_memory_tool_result(result: dict | None) -> list[dict]:
    """Normalize memory MCP tool results into a list of entry dicts."""
    if not result:
        return []

    structured = result.get("structuredContent")
    if isinstance(structured, dict):
        for key in ("memories", "results", "entries", "items"):
            val = structured.get(key)
            if isinstance(val, list):
                return [x for x in val if isinstance(x, dict)]
        if structured.get("title") or structured.get("content"):
            return [structured]

    for block in result.get("content", []):
        if not isinstance(block, dict):
            continue
        text = block.get("text")
        if not text:
            continue
        text = text.strip()
        try:
            parsed = json.loads(text)
        except json.JSONDecodeError:
            continue
        if isinstance(parsed, list):
            return [x for x in parsed if isinstance(x, dict)]
        if isinstance(parsed, dict):
            for key in ("memories", "results", "entries", "items"):
                val = parsed.get(key)
                if isinstance(val, list):
                    return [x for x in val if isinstance(x, dict)]
            if parsed.get("title") or parsed.get("content"):
                return [parsed]
    return []


def _memory_vessel_from_hit(hit: dict) -> str | None:
    for tag in hit.get("tags") or []:
        if isinstance(tag, str) and tag.startswith("vessel:"):
            return tag.split(":", 1)[1].replace("-", " ")
    content = hit.get("content") or ""
    m = re.search(r"^Vessel:\s*(.+)$", content, re.MULTILINE)
    if m:
        return m.group(1).strip()
    title = hit.get("title") or ""
    m = re.search(r"^\[([^\]]+)\]", title)
    if m and m.group(1).lower() not in ("task", "fix"):
        return m.group(1).strip()
    return None


def _obsidian_vessel_from_text(text: str) -> str | None:
    if not text.startswith("---"):
        return None
    end = text.find("\n---", 3)
    if end == -1:
        return None
    front = text[3:end]
    for line in front.splitlines():
        if line.lower().startswith("vessel:"):
            return line.split(":", 1)[1].strip()
    return None


def _memory_hit_from_entry(entry: dict, *, project_id: str) -> dict:
    hit = {
        "project_id": entry.get("project_id") or project_id,
        "title": (entry.get("title") or "").strip(),
        "content": (entry.get("content") or "")[:_EXCERPT_CHARS],
        "memory_type": entry.get("memory_type") or entry.get("type") or "note",
        "importance": entry.get("importance"),
        "tags": entry.get("tags") or [],
        "source": "memory",
    }
    return _enrich_hit_vessel(hit)


def _enrich_hit_vessel(hit: dict, *, vessel_name: str | None = None) -> dict:
    vessel = hit.get("vessel")
    if not vessel and hit.get("source") == "memory":
        vessel = _memory_vessel_from_hit(hit)
    if not vessel and hit.get("source") == "obsidian":
        vessel = _obsidian_vessel_from_text(hit.get("excerpt") or "")
    if vessel:
        hit["vessel"] = vessel
    elif vessel_name:
        hit["vessel"] = vessel_name
    return hit


async def _search_memory_projects(query: str, project_ids: list[str]) -> list[dict]:
    hits: list[dict] = []
    seen_titles: set[str] = set()

    for project_id in project_ids:
        try:
            result = await integrations.memory_mcp_call(
                "memory_search",
                {"project_id": project_id, "query": query, "limit": 5},
            )
            for entry in _parse_memory_tool_result(result):
                hit = _memory_hit_from_entry(entry, project_id=project_id)
                key = hit["title"].lower()
                if not hit["title"] or key in seen_titles:
                    continue
                seen_titles.add(key)
                hits.append(hit)
        except Exception as exc:  # noqa: BLE001
            logger.warning("memory_search %s failed: %s", project_id, exc)

    return hits[:_MAX_MEMORY_HITS]


async def _recent_memory(project_ids: list[str]) -> list[dict]:
    hits: list[dict] = []
    for project_id in project_ids[:2]:
        try:
            result = await integrations.memory_mcp_call(
                "memory_list_recent",
                {"project_id": project_id, "limit": 10},
            )
            for entry in _parse_memory_tool_result(result):
                hits.append(_memory_hit_from_entry(entry, project_id=project_id))
        except Exception as exc:  # noqa: BLE001
            logger.warning("memory_list_recent %s failed: %s", project_id, exc)
    return hits


def _obsidian_search_roots(vault: str) -> list[str]:
    roots: list[str] = []
    for rel in settings.obsidian_search_paths.split(","):
        rel = rel.strip()
        if not rel:
            continue
        path = os.path.join(vault, rel)
        if os.path.isdir(path):
            roots.append(path)
    tasks = os.path.join(vault, settings.obsidian_tasks_folder)
    if os.path.isdir(tasks) and tasks not in roots:
        roots.insert(0, tasks)
    return roots


def _score_obsidian_file(
    path: str, text: str, keywords: list[str], *, vessel_name: str | None = None
) -> int:
    name = os.path.basename(path).lower()
    body = text.lower()
    score = 0
    for kw in keywords:
        if kw in name:
            score += 3
        if kw in body:
            score += 1
    if vessel_name:
        vslug = vessel_name.lower().replace(" ", "-")
        note_vessel = _obsidian_vessel_from_text(text)
        if note_vessel and note_vessel.lower() == vessel_name.lower():
            score += 5
        elif vslug in name or vessel_name.lower() in body:
            score += 2
    return score


def _obsidian_title(text: str, filename: str) -> str:
    for line in text.splitlines():
        if line.startswith("# "):
            return line[2:].strip()
    return os.path.splitext(filename)[0].replace("-", " ")


async def search_obsidian_notes(
    title: str, issue: str, *, vessel_name: str | None = None
) -> list[dict]:
    """Search cloned Obsidian vault for notes related to this issue."""
    vault = await integrations.ensure_vault()
    if not vault:
        return []

    keywords = extract_search_keywords(f"{title} {issue}")
    if vessel_name:
        keywords = keywords + [vessel_name.lower().replace(" ", "-"), vessel_name.lower()]
    if not keywords:
        return []

    # The vault scan is synchronous disk I/O — run it off the event loop.
    return await asyncio.to_thread(_scan_obsidian_notes, vault, keywords, vessel_name)


def _scan_obsidian_notes(
    vault: str, keywords: list[str], vessel_name: str | None
) -> list[dict]:
    scored: list[tuple[int, dict]] = []
    for root in _obsidian_search_roots(vault):
        for dirpath, _, files in os.walk(root):
            for filename in files:
                if not filename.endswith(".md"):
                    continue
                abs_path = os.path.join(dirpath, filename)
                try:
                    with open(abs_path, encoding="utf-8", errors="replace") as fh:
                        text = fh.read()
                except OSError:
                    continue
                score = _score_obsidian_file(abs_path, text, keywords, vessel_name=vessel_name)
                if score <= 0:
                    continue
                rel = os.path.relpath(abs_path, vault)
                excerpt = text[:_EXCERPT_CHARS].strip()
                hit = {
                    "path": rel,
                    "title": _obsidian_title(text, filename),
                    "excerpt": excerpt,
                    "score": score,
                    "source": "obsidian",
                }
                scored.append((score, _enrich_hit_vessel(hit, vessel_name=vessel_name)))

    scored.sort(key=lambda x: (-x[0], x[1]["path"]))
    return [hit for _, hit in scored[:_MAX_OBSIDIAN_HITS]]


async def gather_investigation_context(
    title: str,
    issue: str,
    *,
    vessel_name: str | None = None,
    vessel_id: int | None = None,
) -> dict[str, Any]:
    """Search memory MCP + Obsidian vault for prior related investigations."""
    keywords = extract_search_keywords(f"{title} {issue}")
    query_parts = list(keywords)
    if vessel_name:
        query_parts.insert(0, vessel_name)
    query = " ".join(query_parts) if query_parts else f"{title} {issue}"[:200]

    project_ids = [
        p.strip()
        for p in settings.memory_search_projects.split(",")
        if p.strip()
    ]
    if settings.memory_task_project_id not in project_ids:
        project_ids.insert(0, settings.memory_task_project_id)

    memory_hits = await _search_memory_projects(query, project_ids)

    # If keyword search is sparse, add recent task memories for continuity.
    if len(memory_hits) < 3:
        recent = await _recent_memory(project_ids)
        seen = {h["title"].lower() for h in memory_hits if h.get("title")}
        for hit in recent:
            t = hit.get("title", "").lower()
            if t and t not in seen:
                memory_hits.append(hit)
                seen.add(t)
            if len(memory_hits) >= _MAX_MEMORY_HITS:
                break

    obsidian_hits = await search_obsidian_notes(title, issue, vessel_name=vessel_name)

    return {
        "query": query,
        "keywords": keywords,
        "vessel_name": vessel_name,
        "vessel_id": vessel_id,
        "memory_hits": memory_hits[:_MAX_MEMORY_HITS],
        "obsidian_hits": obsidian_hits,
        "memory_count": len(memory_hits),
        "obsidian_count": len(obsidian_hits),
    }