Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
99 changes: 99 additions & 0 deletions model_capability_check.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,99 @@
"""Ollama model capability verification for autonomous NeuroRift agent mode."""

from __future__ import annotations

import json
import subprocess
from typing import Any

CAPABILITY_PROMPT = """Evaluate whether you can reliably perform the following tasks:

* structured tool invocation
* Linux command generation
* file creation and modification
* interpreting tool outputs
* multi-step reasoning for autonomous agents

Return a JSON response:
{
"tool_usage": true/false,
"command_generation": true/false,
"filesystem_operations": true/false,
"multi_step_reasoning": true/false,
"agent_ready": true/false
}"""


def _extract_json(raw_text: str) -> dict[str, Any]:
raw_text = (raw_text or "").strip()
start = raw_text.find("{")
end = raw_text.rfind("}")
if start == -1 or end == -1 or end <= start:
raise ValueError("No JSON object found in model response")
Comment on lines +28 to +32
Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggestion: The JSON parser currently slices from the first { to the last } in the whole response, which breaks when the model outputs extra brace-containing text before or after the JSON object. This causes valid capability responses to be rejected as invalid JSON. Parse the first decodable JSON object instead of using greedy brace slicing. [logic error]

Severity Level: Major ⚠️
-`run-agent` startup blocked by false JSON parse failure.
- ⚠️ Capability check emits misleading `invalid_capability_json` errors.
Suggested change
raw_text = (raw_text or "").strip()
start = raw_text.find("{")
end = raw_text.rfind("}")
if start == -1 or end == -1 or end <= start:
raise ValueError("No JSON object found in model response")
decoder = json.JSONDecoder()
for i, ch in enumerate(raw_text):
if ch != "{":
continue
try:
obj, _ = decoder.raw_decode(raw_text[i:])
except json.JSONDecodeError:
continue
if isinstance(obj, dict):
return obj
raise ValueError("No JSON object found in model response")
Steps of Reproduction ✅
1. Run NeuroRift through the real agent entrypoint `run-agent` (subcommand defined at
`neurorift_main.py:937`), which reaches model checking in `_async_main` at
`neurorift_main.py:1007-1019`.

2. Use a model/runner output that includes extra brace text around valid JSON (the code
path is `verify_model_capabilities()` in `model_capability_check.py:35`, called at
`neurorift_main.py:1018`).

3. `_extract_json()` at `model_capability_check.py:28-32` slices from first `{` to last
`}`, so mixed text with additional braces creates an invalid combined JSON slice.

4. `json.loads(...)` fails, `verify_model_capabilities()` returns
`invalid_capability_json` at `model_capability_check.py:62`, and `run-agent` is blocked by
the readiness gate at `neurorift_main.py:1019-1023`.
Prompt for AI Agent 🤖
This is a comment left during a code review.

**Path:** model_capability_check.py
**Line:** 28:32
**Comment:**
	*Logic Error: The JSON parser currently slices from the first `{` to the last `}` in the whole response, which breaks when the model outputs extra brace-containing text before or after the JSON object. This causes valid capability responses to be rejected as invalid JSON. Parse the first decodable JSON object instead of using greedy brace slicing.

Validate the correctness of the flagged issue. If correct, How can I resolve this? If you propose a fix, implement it and please make it concise.
👍 | 👎

return json.loads(raw_text[start : end + 1])


def verify_model_capabilities(model_name: str) -> dict[str, Any]:
try:
proc = subprocess.run(
["ollama", "run", model_name, CAPABILITY_PROMPT],
capture_output=True,
text=True,
timeout=120,
check=False,
)
except FileNotFoundError:
return {"ok": False, "error": "ollama_missing", "agent_ready": False}
except Exception as exc: # defensive for unstable environments
return {
"ok": False,
"error": f"ollama_exec_error:{type(exc).__name__}",
"agent_ready": False,
}

if proc.returncode != 0:
return {
"ok": False,
"error": f"ollama_returned_{proc.returncode}",
"stderr": proc.stderr,
"agent_ready": False,
}

try:
parsed = _extract_json(proc.stdout)
except Exception as exc:
return {
"ok": False,
"error": f"invalid_capability_json:{type(exc).__name__}",
"raw": proc.stdout[:1000],
"agent_ready": False,
}

required = {
"tool_usage",
"command_generation",
"filesystem_operations",
"multi_step_reasoning",
"agent_ready",
}
if not required.issubset(parsed):
return {
"ok": False,
"error": "capability_fields_missing",
"parsed": parsed,
Comment on lines +82 to +83
Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggestion: agent_ready is used as a truthy value without enforcing boolean type, so a model returning "false" (string) is treated as ready by callers because non-empty strings are truthy. Normalize agent_ready to a strict boolean before returning. [logic error]

Severity Level: Critical 🚨
-`run-agent` readiness gate bypassed by string booleans.
- ⚠️ Autonomous mode may run with unsupported models.
Suggested change
"error": "capability_fields_missing",
"parsed": parsed,
parsed["agent_ready"] = parsed.get("agent_ready") is True
parsed["ok"] = parsed["agent_ready"]
return parsed
Steps of Reproduction ✅
1. Start from the real gate path: `run-agent` in `_async_main`
(`neurorift_main.py:1007-1023`) calls `verify_model_capabilities()` at
`neurorift_main.py:1018`.

2. Return a capability JSON object where required keys exist but `agent_ready` is a
string, e.g. `"agent_ready": "false"` (parsed in `model_capability_check.py:58`, required
fields validated at `model_capability_check.py:67-80`).

3. Current code sets `parsed["ok"] = bool(parsed.get("agent_ready"))`
(`model_capability_check.py:82`), which converts non-empty `"false"` to `True`.

4. `run-agent` then checks `if not capability.get("agent_ready")`
(`neurorift_main.py:1019`); string `"false"` is truthy, so the failure branch is skipped
and agent runtime proceeds incorrectly.
Prompt for AI Agent 🤖
This is a comment left during a code review.

**Path:** model_capability_check.py
**Line:** 82:83
**Comment:**
	*Logic Error: `agent_ready` is used as a truthy value without enforcing boolean type, so a model returning `"false"` (string) is treated as ready by callers because non-empty strings are truthy. Normalize `agent_ready` to a strict boolean before returning.

Validate the correctness of the flagged issue. If correct, How can I resolve this? If you propose a fix, implement it and please make it concise.
👍 | 👎

"agent_ready": False,
}

parsed["ok"] = bool(parsed.get("agent_ready"))
return parsed


if __name__ == "__main__":
import argparse

parser = argparse.ArgumentParser(
description="Check Ollama model capability for NeuroRift agent mode"
)
parser.add_argument("--model", required=True)
args = parser.parse_args()
print(json.dumps(verify_model_capabilities(args.model), indent=2))
1 change: 1 addition & 0 deletions neurorift/__init__.py
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@

1 change: 1 addition & 0 deletions neurorift/channels/__init__.py
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@

3 changes: 3 additions & 0 deletions neurorift/channels/channel_router.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,3 @@
class ChannelRouter:
def route(self, msg: dict):
return msg
12 changes: 12 additions & 0 deletions neurorift/channels/cli_channel.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,12 @@
class Channel:
def connect(self):
return True

def receive_message(self, payload=None):
return payload or {}

def send_message(self, message):
return message

def close(self):
return True
12 changes: 12 additions & 0 deletions neurorift/channels/discord_channel.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,12 @@
class Channel:
def connect(self):
return True

def receive_message(self, payload=None):
return payload or {}

def send_message(self, message):
return message

def close(self):
return True
12 changes: 12 additions & 0 deletions neurorift/channels/telegram_channel.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,12 @@
class Channel:
def connect(self):
return True

def receive_message(self, payload=None):
return payload or {}

def send_message(self, message):
return message

def close(self):
return True
12 changes: 12 additions & 0 deletions neurorift/channels/web_channel.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,12 @@
class Channel:
def connect(self):
return True

def receive_message(self, payload=None):
return payload or {}

def send_message(self, message):
return message

def close(self):
return True
1 change: 1 addition & 0 deletions neurorift/clawhub/__init__.py
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@

2 changes: 2 additions & 0 deletions neurorift/clawhub/clawhub_api.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,2 @@
class ClawHubAPI:
base_url = "https://clawhub.example/api"
18 changes: 18 additions & 0 deletions neurorift/clawhub/clawhub_client.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,18 @@
from pathlib import Path
import shutil


class ClawHubClient:
def __init__(self, cache_dir: Path):
self.cache_dir = cache_dir
cache_dir.mkdir(parents=True, exist_ok=True)

def fetch_skill(self, skill_name: str, source_dir: Path) -> Path:
src = source_dir / skill_name
if not src.exists():
raise FileNotFoundError(skill_name)
Comment on lines +12 to +13
Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggestion: The existence check accepts regular files, but shutil.copytree only works on directories; if a file exists with that name, this path raises NotADirectoryError at runtime. Validate that the source is a directory. [possible bug]

Severity Level: Major ⚠️
-`--clawhub` install crashes on file-like skill names.
- ⚠️ User-facing skill install flow becomes unreliable.
Suggested change
if not src.exists():
raise FileNotFoundError(skill_name)
if not src.is_dir():
raise FileNotFoundError(skill_name)
Steps of Reproduction ✅
1. Execute `python3 /workspace/NeuroRift/neurorift_main.py --clawhub __init__.py`;
`--clawhub` input is accepted by parser at `neurorift_main.py:914` and routed to
`install_clawhub()` at `neurorift_main.py:982-983`.

2. `SkillManager.install_clawhub()` forwards `__init__.py` directly to `fetch_skill()` at
`neurorift/skills/skill_manager.py:13-14`.

3. In `fetch_skill()`, existence-only check passes (`src.exists()`) at
`neurorift/clawhub/clawhub_client.py:7` because
`/workspace/NeuroRift/neurorift/skill_store/installed/__init__.py` is a real file
(verified in `neurorift/skill_store/installed` listing).

4. `shutil.copytree(src, dst)` at `neurorift/clawhub/clawhub_client.py:10` then raises
`NotADirectoryError`, aborting the install path.
Prompt for AI Agent 🤖
This is a comment left during a code review.

**Path:** neurorift/clawhub/clawhub_client.py
**Line:** 12:13
**Comment:**
	*Possible Bug: The existence check accepts regular files, but `shutil.copytree` only works on directories; if a file exists with that name, this path raises `NotADirectoryError` at runtime. Validate that the source is a directory.

Validate the correctness of the flagged issue. If correct, How can I resolve this? If you propose a fix, implement it and please make it concise.
👍 | 👎

dst = self.cache_dir / skill_name
if dst.exists():
shutil.rmtree(dst)
shutil.copytree(src, dst)
Comment on lines +10 to +17
Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

⚠️ Potential issue | 🔴 Critical

Reject traversal in skill_name before touching the filesystem.

skill_name is joined directly into both source_dir and cache_dir. Values like ../foo can make Line 16 delete outside the cache root and Line 17 copy arbitrary directories into it.

Suggested fix
     def fetch_skill(self, skill_name: str, source_dir: Path) -> Path:
-        src = source_dir / skill_name
-        if not src.exists():
+        src_root = source_dir.resolve()
+        dst_root = self.cache_dir.resolve()
+        src = (src_root / skill_name).resolve()
+        dst = (dst_root / skill_name).resolve()
+        if src_root not in src.parents or dst_root not in dst.parents:
+            raise ValueError("invalid skill name")
+        if not src.is_dir():
             raise FileNotFoundError(skill_name)
-        dst = self.cache_dir / skill_name
         if dst.exists():
             shutil.rmtree(dst)
         shutil.copytree(src, dst)
         return dst
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.

In `@neurorift/clawhub/clawhub_client.py` around lines 10 - 17, Reject path
traversal in fetch_skill by validating skill_name before any filesystem
operations: ensure skill_name is a single path segment (no "..", no absolute
paths, no path separators) and normalize/resolve only after validation; use the
validated name to build src = source_dir / skill_name and dst = self.cache_dir /
skill_name, then proceed to check existence and perform
shutil.rmtree/shutil.copytree. Refer to symbols fetch_skill, skill_name,
source_dir, cache_dir, src, and dst when making the change.

return dst
3 changes: 3 additions & 0 deletions neurorift/clawhub/clawhub_resolver.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,3 @@
class ClawHubResolver:
def resolve(self, skill_name: str) -> str:
return f"https://clawhub.example/skills/{skill_name}"
1 change: 1 addition & 0 deletions neurorift/cli/__init__.py
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@

7 changes: 7 additions & 0 deletions neurorift/cli/neurorift_cli.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,7 @@
"""Package-level CLI helper that delegates to global entrypoint."""

from __future__ import annotations

from neurorift_cli import build_parser, install_global_wrapper, main

__all__ = ["build_parser", "install_global_wrapper", "main"]
1 change: 1 addition & 0 deletions neurorift/config/__init__.py
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@

1 change: 1 addition & 0 deletions neurorift/config/channel_config.py
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
CHANNELS = ["cli", "websocket", "discord", "telegram", "api"]
1 change: 1 addition & 0 deletions neurorift/config/model_config.py
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
MODEL_PROVIDERS = ["deepseek", "mistral", "llama", "openai"]
7 changes: 7 additions & 0 deletions neurorift/config/settings.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,7 @@
from dataclasses import dataclass
from pathlib import Path


@dataclass
class Settings:
data_dir: Path = Path.home() / ".neurorift"
1 change: 1 addition & 0 deletions neurorift/core/__init__.py
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@

3 changes: 3 additions & 0 deletions neurorift/core/agent_loop.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,3 @@
class AgentLoop:
async def handle_message(self, session, message: str) -> dict:
return {"success": True, "response": f"Session {session.session_id}: {message}"}
17 changes: 17 additions & 0 deletions neurorift/core/agent_manager.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,17 @@
from dataclasses import dataclass
from neurorift.core.agent_loop import AgentLoop
from neurorift.sessions.session_manager import SessionManager


@dataclass
class AgentHandle:
session_id: str
user_id: str
channel: str


class AgentManager:
def __init__(self, session_manager: SessionManager, agent_loop: AgentLoop):
self.session_manager = session_manager
self.agent_loop = agent_loop
self.handles: dict[str, AgentHandle] = {}
16 changes: 16 additions & 0 deletions neurorift/core/planner.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,16 @@
from dataclasses import dataclass


@dataclass
class PlannedStep:
order: int
action: str


class Planner:
def create_plan(self, objective: str) -> list[PlannedStep]:
return (
[PlannedStep(order=1, action=f"Investigate: {objective}")]
if objective
else []
)
3 changes: 3 additions & 0 deletions neurorift/core/task_router.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,3 @@
class TaskRouter:
def route(self, task: str) -> dict:
return {"task": task, "route": "agent_loop"}
1 change: 1 addition & 0 deletions neurorift/execution/__init__.py
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@

12 changes: 12 additions & 0 deletions neurorift/execution/command_runner.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,12 @@
import subprocess


class CommandRunner:
def run(self, cmd: list[str]):
p = subprocess.run(cmd, capture_output=True, text=True, check=False)
Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggestion: If the executable in the command does not exist, subprocess.run raises FileNotFoundError and the method crashes. Catch this exception and return a normal error payload so callers can handle missing tools safely. [possible bug]

Severity Level: Major ⚠️
- ⚠️ Future execution APIs crash on missing binaries.
- ⚠️ Error handling contract breaks for CommandRunner callers.
Suggested change
p = subprocess.run(cmd, capture_output=True, text=True, check=False)
try:
p = subprocess.run(cmd, capture_output=True, text=True, check=False)
except FileNotFoundError as exc:
return {
"success": False,
"stdout": "",
"stderr": str(exc),
"exit_code": 127,
}
Steps of Reproduction ✅
1. Confirm this runner is currently scaffolded but callable:
`neurorift/execution/command_runner.py:2-5`; no internal caller found except subclass
declaration `neurorift/execution/sandbox_runner.py:1-3`.

2. Execute `from neurorift.execution.command_runner import CommandRunner;
CommandRunner().run(["definitely_not_installed_bin"])`.

3. Call reaches `subprocess.run(...)` at `command_runner.py:4` with nonexistent
executable.

4. Observe uncaught `FileNotFoundError`, so no `{success, stderr, exit_code}` response is
returned.
Prompt for AI Agent 🤖
This is a comment left during a code review.

**Path:** neurorift/execution/command_runner.py
**Line:** 6:6
**Comment:**
	*Possible Bug: If the executable in the command does not exist, `subprocess.run` raises `FileNotFoundError` and the method crashes. Catch this exception and return a normal error payload so callers can handle missing tools safely.

Validate the correctness of the flagged issue. If correct, How can I resolve this? If you propose a fix, implement it and please make it concise.
👍 | 👎

return {
"success": p.returncode == 0,
"stdout": p.stdout,
"stderr": p.stderr,
"exit_code": p.returncode,
}
8 changes: 8 additions & 0 deletions neurorift/execution/resource_limits.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,8 @@
from dataclasses import dataclass


@dataclass
class ResourceLimits:
timeout_seconds: int = 30
memory_limit_mb: int = 512
cpu_seconds: int = 20
9 changes: 9 additions & 0 deletions neurorift/execution/retry_manager.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,9 @@
import time


class RetryManager:
def __init__(self, retries=2):
self.retries = retries

def wait(self, attempt):
time.sleep(2**attempt)
5 changes: 5 additions & 0 deletions neurorift/execution/sandbox_runner.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,5 @@
from neurorift.execution.command_runner import CommandRunner


class SandboxRunner(CommandRunner):
pass
1 change: 1 addition & 0 deletions neurorift/gateway/__init__.py
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@

3 changes: 3 additions & 0 deletions neurorift/gateway/api_server.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,3 @@
class APIServer:
def start(self):
return True
3 changes: 3 additions & 0 deletions neurorift/gateway/auth_manager.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,3 @@
class AuthManager:
def validate(self, token: str):
return bool(token)
Comment on lines +2 to +3
Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

⚠️ Potential issue | 🔴 Critical

AuthManager.validate() is effectively an auth bypass for any non-empty token.

Line [3] accepts arbitrary non-empty input as valid credentials. This is a critical auth flaw; validation must be cryptographic/authoritative and fail closed on errors.

Safer pattern (inject verifier, fail closed)
 class AuthManager:
-    def validate(self, token: str):
-        return bool(token)
+    def __init__(self, token_verifier):
+        self._token_verifier = token_verifier
+
+    def validate(self, token: str) -> bool:
+        if not isinstance(token, str) or not token.strip():
+            return False
+        try:
+            return bool(self._token_verifier(token))
+        except Exception:
+            return False
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.

In `@neurorift/gateway/auth_manager.py` around lines 2 - 3, AuthManager.validate
currently treats any non-empty string as valid; replace the bool(token) bypass
with a real verification call by injecting or using an authoritative verifier
(e.g., a provided token_verifier.verify or verify_token method) inside
AuthManager.validate (or accept a verifier in __init__), call that verifier with
the token, and return its boolean result; wrap the call in try/except and always
fail closed (return False) on exceptions or if no verifier is configured to
ensure invalid/errored tokens are rejected.

3 changes: 3 additions & 0 deletions neurorift/gateway/websocket_gateway.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,3 @@
class WebSocketGateway:
def start(self):
return True
1 change: 1 addition & 0 deletions neurorift/memory/__init__.py
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@

6 changes: 6 additions & 0 deletions neurorift/memory/long_term_memory.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,6 @@
class LongTermMemory:
def __init__(self):
self.data = {}

def store(self, user_id: str, key: str, value):
self.data.setdefault(user_id, {})[key] = value
3 changes: 3 additions & 0 deletions neurorift/memory/memory_compaction.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,3 @@
class MemoryCompaction:
def compact(self, messages: list[str]):
return {"summary": " ".join(messages[:10]), "kept": messages[-10:]}
9 changes: 9 additions & 0 deletions neurorift/memory/short_term_memory.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,9 @@
from collections import defaultdict, deque


class ShortTermMemory:
def __init__(self, limit: int = 30):
self.buffers = defaultdict(lambda: deque(maxlen=limit))

def add(self, sid: str, msg: str):
self.buffers[sid].append(msg)
Comment on lines +5 to +9
Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggestion: buffers is bounded per session but unbounded in number of sessions, so repeatedly adding messages for new sid values grows memory forever and can eventually exhaust process memory. Add a cap on tracked sessions and evict the oldest session buffer when that cap is reached. [resource leak]

Severity Level: Major ⚠️
- ⚠️ Long-running agents accumulate per-session buffers indefinitely.
- ⚠️ neurorift memory module risks process RAM growth.
Suggested change
def __init__(self, limit: int = 30):
self.buffers = defaultdict(lambda: deque(maxlen=limit))
def add(self, sid: str, msg: str):
self.buffers[sid].append(msg)
def __init__(self, limit: int = 30, max_sessions: int = 1000):
self.buffers = defaultdict(lambda: deque(maxlen=limit))
self.max_sessions = max_sessions
self._session_order = deque()
def add(self, sid: str, msg: str):
if sid not in self.buffers:
if len(self._session_order) >= self.max_sessions:
oldest_sid = self._session_order.popleft()
self.buffers.pop(oldest_sid, None)
self._session_order.append(sid)
self.buffers[sid].append(msg)
Steps of Reproduction ✅
1. Verify exposure: `setup.py:6` packages `neurorift.*`, so
`neurorift.memory.short_term_memory.ShortTermMemory` is importable API.

2. Verify current implementation: `neurorift/memory/short_term_memory.py:3-4` stores
messages in `self.buffers[sid]` with no session-count guard.

3. Use realistic session IDs from `neurorift/sessions/session_manager.py:10-12`
(`uuid4`-style `sid`) and repeatedly call `ShortTermMemory.add(sid, msg)` with new IDs.

4. Observe unbounded key growth (`len(memory.buffers)` keeps increasing) because each new
`sid` creates a new deque and no eviction path exists.
Prompt for AI Agent 🤖
This is a comment left during a code review.

**Path:** neurorift/memory/short_term_memory.py
**Line:** 5:9
**Comment:**
	*Resource Leak: `buffers` is bounded per session but unbounded in number of sessions, so repeatedly adding messages for new `sid` values grows memory forever and can eventually exhaust process memory. Add a cap on tracked sessions and evict the oldest session buffer when that cap is reached.

Validate the correctness of the flagged issue. If correct, How can I resolve this? If you propose a fix, implement it and please make it concise.
👍 | 👎

6 changes: 6 additions & 0 deletions neurorift/memory/vector_memory.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,6 @@
class VectorMemory:
def semantic_search(self, user_id: str, query: str, top_k: int = 5):
return []

def time_based_retrieval(self, user_id: str, limit: int = 10):
return []
1 change: 1 addition & 0 deletions neurorift/models/__init__.py
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@

3 changes: 3 additions & 0 deletions neurorift/models/context_builder.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,3 @@
class ContextBuilder:
def build(self, session, memories):
return {"session": session.session_id, "memories": memories}
3 changes: 3 additions & 0 deletions neurorift/models/model_failover.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,3 @@
class ModelFailover:
def pick(self, providers):
return providers[0] if providers else "fallback"
8 changes: 8 additions & 0 deletions neurorift/models/model_router.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,8 @@
class ModelRouter:
async def generate(self, prompt: str):
return {
"response": prompt,
"model": "openai",
"tokens": len(prompt.split()),
"cost": 0.0,
}
3 changes: 3 additions & 0 deletions neurorift/models/prompt_builder.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,3 @@
class PromptBuilder:
def build(self, message: str, context: dict):
return f"{context}\n{message}"
1 change: 1 addition & 0 deletions neurorift/sessions/__init__.py
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@

29 changes: 29 additions & 0 deletions neurorift/sessions/session_context.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,29 @@
from dataclasses import dataclass, field
from enum import Enum
from datetime import datetime, timezone


class SessionState(str, Enum):
CREATED = "CREATED"
ACTIVE = "ACTIVE"
IDLE = "IDLE"
PAUSED = "PAUSED"
CLOSED = "CLOSED"


@dataclass
class SessionContext:
session_id: str
user_id: str
channel: str
state: SessionState = SessionState.CREATED
Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggestion: SessionStore.load() rebuilds this dataclass from JSON, where state comes back as a plain string. Without normalization, the object violates its own contract (SessionState) and will break when enum behavior is expected (for example accessing enum attributes). Coerce string input to SessionState in __post_init__. [type error]

Severity Level: Major ⚠️
- ⚠️ Reloaded sessions violate `SessionContext.state` enum contract.
- ❌ Enum attribute access crashes after session reload.
- ⚠️ Future state-driven logic becomes type-unsafe.
Suggested change
state: SessionState = SessionState.CREATED
state: SessionState = SessionState.CREATED
def __post_init__(self) -> None:
if isinstance(self.state, str):
self.state = SessionState(self.state)
Steps of Reproduction ✅
1. Create and persist a session through `SessionManager.create_session()` at
`neurorift/sessions/session_manager.py:10-12`; this calls `SessionStore.save()` at
`neurorift/sessions/session_store.py:9`.

2. Simulate a new process (new `SessionManager` instance) and call
`SessionManager.get_session()` at `neurorift/sessions/session_manager.py:13`; this falls
through to `SessionStore.load()` at `neurorift/sessions/session_store.py:10-12`.

3. `SessionStore.load()` reconstructs via `SessionContext(**json.loads(...))`
(`neurorift/sessions/session_store.py:12`), while `SessionContext` has no `__post_init__`
normalization (`neurorift/sessions/session_context.py:8-19`), so `state` remains plain
`str`.

4. Access enum semantics on the loaded object (for example `loaded.state.name`) and
observe runtime failure: `AttributeError: 'str' object has no attribute 'name'` (verified
by executing this exact call chain with `PYTHONPATH=/workspace/NeuroRift`).
Prompt for AI Agent 🤖
This is a comment left during a code review.

**Path:** neurorift/sessions/session_context.py
**Line:** 19:19
**Comment:**
	*Type Error: `SessionStore.load()` rebuilds this dataclass from JSON, where `state` comes back as a plain string. Without normalization, the object violates its own contract (`SessionState`) and will break when enum behavior is expected (for example accessing enum attributes). Coerce string input to `SessionState` in `__post_init__`.

Validate the correctness of the flagged issue. If correct, How can I resolve this? If you propose a fix, implement it and please make it concise.
👍 | 👎

message_history: list[dict] = field(default_factory=list)
tool_usage: list[dict] = field(default_factory=list)
context_window: list[str] = field(default_factory=list)
memory_references: list[str] = field(default_factory=list)
created_at: str = field(
default_factory=lambda: datetime.now(timezone.utc).isoformat()
)
updated_at: str = field(
default_factory=lambda: datetime.now(timezone.utc).isoformat()
)
Loading