Security audit
Security checks for vulnerabilities and agentic risk
This game needs review because narration can send text online by default without clear notice, and restarting can overwrite an existing save without a backup.
Create games with --tts off, or disable narration using a meta action with tts set to false, unless you accept online speech processing. Use explicit game commands, unique save IDs and independent backups. Starting again with an existing ID replaces its save without creating a new backup. Opening the orbit HTML online also attempts to load three.js from unpkg.
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.
def test_path_traversal_user_id_is_rejected(self):
with self.assertRaises(ValueError):
new_state(user_id="../etc/passwd")
def test_two_calls_with_the_same_seed_share_a_mole_assignment(self):
a = new_state(user_id="a1", seed=99)
Suspicious Unicode normalization or mixed-script content
The document describes optional speech generation using edge-tts without explaining that narration text is transmitted to Microsoft's online speech service. If narration incorporates player-provided names or other private text, enabling this feature could disclose that information outside the local environment without informed consent; its optional, fictional-game context reduces the likely exposure.
Suspicious Unicode normalization or mixed-script content
Suspicious Unicode normalization or mixed-script content
Suspicious Unicode normalization or mixed-script content
Without declared permissions the skill's intent is opaque and cannot be validated.
The workflow references edge-tts and online neural voices without explaining that speech generation sends narration to an external service. Enabling audio is not necessarily informed consent to that disclosure, and narration may contain player-provided names or other personal details; the enabled-state requirement and restriction to short game narration limit the exposure.
Suspicious Unicode normalization or mixed-script content
edge-tts sends the supplied narration text to an external speech service, while this file provides no disclosure or consent check and treats TTS as enabled when the setting is absent. Game narration is ordinarily low sensitivity, but user-supplied or private information included in the narration would also leave the local environment; the 4,000-character limit does not prevent that disclosure.
subprocess module calls execute external commands. Without careful input validation, this enables command injection.
out = Path(args.out) if args.out else Path.home() / ".realmarsshot" / "tts" / f"{args.user_id}-{datetime.now().strftime('%Y%m%d-%H%M%S')}.mp3"
out.parent.mkdir(parents=True, exist_ok=True)
text = args.text.strip()[:4000]
proc = subprocess.run(
["edge-tts", "--voice", voice, "--text", text, "--write-media", str(out)],
capture_output=True,
text=True,
evaluate_month() determines survival from local production without considering stored food, oxygen, or water, and apply_month() calls _harm() on that basis before evaluating store-supported foothold survival. Consequently, a trial crew with adequate stored supplies can suffer unwarranted deaths, population-clock resets, and financial penalties. This is a simulation-integrity defect; the game context limits the demonstrated consequences to game state rather than real-world harm or host compromise.
For version-1 saves with achieved biology or machine footholds, migration raises the corresponding simulated years to the configured win threshold, contrary to its documented promise not to grant deep-time years. This can bypass progression requirements in downstream checks that trust those counters, although the demonstrated impact is limited to game-state integrity rather than host security or sensitive-data exposure.
At L228–L229, write() creates parent directories and writes to the supplied path, replacing existing contents if the file already exists. There is no confirmation, log message, docstring, or comment explaining these filesystem effects.
The ordinary word “火种” can appear in conversation without expressing an intent to start this game, yet the workflow directs the agent to create a game after activation. This can cause unwanted tool execution and persistent state changes; the documented reuse of a default user ID also creates a risk of overwriting an existing game, although adjacent backups reduce the potential damage.
No suspicious patterns detected.