Improve security: redact sensitive evidence, tighten scope checks, secure persistence, and test coverage

[prithvee07]

Motivation

• Prevent secrets from leaking into logs, AI prompts, reports, or persistent state by centrally redacting common credential patterns and ensuring commands stored are sanitized.
• Reduce attack surface from unsafe file permissions and TOCTOU windows by making session and report files/directories private and re-checking scope before executing external tools.
• Harden XML parsing and scope validation to mitigate entity expansion, oversized input, DNS-rebinding, and special-address bypasses.

Description

• Added utils/redaction.py and integrated redact_sensitive_text in core/workflow.py and when saving tool executions so raw tool evidence and analyst inputs have common credential patterns redacted.
• Updated tools/base_tool.py so the logged/stored command is always the sanitized form returned by _sanitize_command_for_logging and never the raw executable arguments.
• Hardened persistence: PentestMemory.save_state and workflow report/session save paths now create directories with mode 0o700 and create files with 0o600 via os.open/os.fchmod to avoid world-readable secrets.
• Re-validate scope immediately before each tool execution and AI-decided tool run in core/workflow.py to reduce TOCTOU/DNS-rebinding windows.
• Tightened utils/scope_validator.py blacklist behavior to block non-global addresses (including IPv6 ULA and IPv4 link-local) and explicit special names, and continue to resolve hostnames and check resolved IPs against blacklist networks.
• Improved tools/nmap.py to reject XML inputs containing DTDs/entities and inputs larger than 10 MiB before parsing to mitigate entity expansion and memory exhaustion.
• Constrained direct dependency upper bounds in pyproject.toml to avoid inadvertent major-version upgrades in release builds.
• Added and updated tests: new tests/test_redaction.py, additional assertions in tests/test_memory.py, new nmap safety tests in tests/test_nmap_tool.py, added behavior test in tests/test_base_tool.py, and extended blacklist cases in tests/test_scope_validator.py.

Testing

• Ran the automated test suite with pytest including tests/test_memory.py, tests/test_redaction.py, tests/test_nmap_tool.py, tests/test_base_tool.py, and tests/test_scope_validator.py, and all tests passed.
• Confirmed new unit tests validate redaction, file permission semantics (private state files/dirs), nmap XML rejection heuristics, sanitized command exposure, and blacklist behavior.

---

Codex Task