Do not open a public issue. Instead, report vulnerabilities privately:

• Email: security@dscode.dev (or open a private security advisory on GitHub)

You will receive a response within 48 hours. We aim to patch confirmed vulnerabilities within 7 days.

• Description of the vulnerability
• Steps to reproduce
• Affected components/versions
• Potential impact
• Suggested fix (if available)

DS Code executes AI-generated tool calls on your system. Our security model has multiple layers:

• Static Blacklist: blocks rm -rf /, mkfs.*, dd if=, fork bombs (:(){ :|:& };:), chmod -R 777 /, sudo rm, sudo mv
• Regex Guard: SafetyGuard applies user-configurable regex patterns to block additional dangerous commands
• Process Isolation: commands run in isolated process groups, killed via kill(-pid, SIGKILL) on drop
• Timeout: default 120s, configurable, hard cap at 600s
• Output Limit: 10MB max combined stdout+stderr

• Path Traversal Prevention: all paths are canonicalized and checked to be within the working directory
• Symlink Resolution: symbolic links are resolved before boundary checks
• Atomic Edits: do_file_edit requires exact unique match, preventing unintended modifications
• Non-Existent Path Safety: resolve_safe_path handles paths to files that don't exist yet

• Stored in ~/.dscode/config.toml
• File permissions should be 600 (owner read/write only)
• Known limitation: keys are stored in plaintext. System keyring integration is planned (see #P2-14).

• MCP servers run as child processes over stdio
• 30-second initialization timeout
• Tool discovery is explicit (tools/list handshake)
• Only tools returned by the server are callable

• Compression pipeline prevents token overflow attacks
• L1–L4 escalation: truncation → summarization → wiki ingestion → hard cutoff
• Tool-call ↔ tool-result chain alignment is preserved during compression

1. File Permissions: chmod 600 ~/.dscode/config.toml
2. API Keys: use read-only API keys with minimal scope when possible
3. Working Directory: run DS Code from the specific project directory, not from / or ~
4. Command Whitelist: for production environments, configure safety.allowed_commands in config
5. Review Tool Calls: the TUI and Desktop UI display tool calls before execution — review them
6. Keep Updated: always run the latest version for security patches

The codebase has been manually audited for:

• Command injection vulnerabilities
• Path traversal attacks
• Unsafe Rust usage (1 instance, justified: Unix kill(-pid, SIGKILL))
• Dependency license compliance (0 GPL/AGPL risks)

(Empty — be the first!)

Last updated: 2025-07-09