Only the latest minor release receives security patches. Older versions are best-effort.

python_maithili executes user-supplied .dmai files. The sandbox in maithili_dsl/cli.py is the project's primary trust boundary. It enforces:

1. Import whitelist. Only modules present in MAITHILI_MODULES (the Maithili → Python module map) are importable from a .dmai file. Raw Python import statements that were not translated from Maithili names are rejected by _validate_imports before exec runs. The __import__ builtin inside the sandbox is replaced with a wrapper that re-checks the whitelist at runtime.
2. Safe builtins. The execution namespace's __builtins__ is a curated dict (_SAFE_BUILTIN_NAMES). Notably absent: exec, eval, compile, open, breakpoint. Any .dmai script that calls these gets a NameError translated to Maithili.
3. No shared globals. exec runs with a fresh globals dict — it cannot read or modify the CLI's own process state.

If you find a way to bypass any of these, please report it.

Please do not open a public issue for security problems. Instead:

• Email: jha.bishwas@gmail.com with the subject line [SECURITY] python_maithili and a proof-of-concept.
• Or use GitHub's private vulnerability reporting at https://github.com/alphacrack/python-maithili-dsl/security/advisories/new

Please include:

• A .dmai file that demonstrates the issue.
• The Python version and OS.
• What the sandbox should have prevented, and what actually happened.

• Acknowledgement: within 3 business days.
• Triage + severity classification: within 7 business days.
• Patch + coordinated disclosure: depends on severity; typically within 30 days for HIGH/CRITICAL findings.

In scope:

• Sandbox escape from within a .dmai file.
• Ability to read or modify files outside the CWD without being invoked by an explicitly-whitelisted module call.
• Denial of service via a small .dmai input (e.g., catastrophic regex backtracking in the transpiler).

Out of scope:

• Issues that require first compromising the machine running python_maithili (e.g., modifying the installed package).
• Social engineering.
• Resource-exhaustion attacks that require arbitrarily large input (basic DoS via huge files).