rea

Least-privilege permissions

• Restrict file system access to explicit target binaries and their parent directories only; deny recursive traversal or access to /home, /root, C:\Users, or credential stores
• Use read-only mode for decompilation and analysis; revoke write/execute permissions on all filesystem operations
• Limit decompiler/analysis tool invocations to sandboxed temporary directories; prevent tool access to system PATH or library directories
• Require explicit allowlist of approved binary hashes or paths; reject analysis requests on unlisted targets

Network isolation

• Disable all outbound network access; if updates to analysis tools are needed, use air-gapped or pre-approved channels only
• Block agent from making HTTP/DNS requests during analysis to prevent exfiltration or C2 callbacks embedded in analyzed code
• Route any legitimate external queries (e.g., symbol lookups) through a centralized proxy with strict URL allowlisting

Monitoring

• Log all binary/source files accessed with timestamps, file hashes, and data volume read; alert on access outside declared scope
• Track decompiler output size and content patterns; flag suspicious exports (credentials, URLs, API keys) for manual review
• Monitor resource consumption (CPU, memory, disk I/O); alert on anomalous spikes indicating unintended deep introspection
• Audit agent reasoning chain and tool calls; detect prompt injection attempts via unusual decompiled code patterns (e.g., embedded instructions)

Red flags to watch for

• Requests to analyze third-party or closed-source commercial software without documented authorization and legal review
• Attempts to access system binaries, interpreters, or security tools outside explicit reverse engineering scope
• Decompiled output containing plaintext secrets, API keys, or database credentials being returned or logged without sanitization
• Agent queries or reasoning that deviate from code analysis into system reconnaissance (process lists, network sockets, user enumeration)