Rea Agent Skill by mytechnotalent

FREE Reverse Engineering Self-Study Course HERE

<br>

Reverse Engineering Assistant (REA) v7.1.0

The reverse-engineering framework you can run anywhere, with nothing installed, and actually trust.

rea suspect

Writes re-suspect.md, re-suspect.pdf, and re-suspect.json next to the target. That's the whole workflow.

Targets in. Answers out. One command. Hybrid autonomous multi-agent reasoning and a deterministic engine. Built in 100% pure Rust — run anywhere, install nothing, and every claim verified.

<br>

New in v7.1.0

REA v7.1.0 adds more artifacts and deeper provenance — still pure Rust, offline, and evidence-backed.

• Model artifacts. GGUF and safetensors files are parsed for version, architecture/name, tensor declarations, data types, and total parameter counts, with a new rea model <file> command and MCP inspect_model_artifact tool.
• Godot PCK. Standalone and embedded Godot resource packs (formats v1–v3, including the trailing locator) are inventoried with paths, sizes, and encryption flags, with a new rea godot <file> command and MCP inspect_godot tool.
• Deeper provenance. Rust cargo-auditable metadata now includes each crate's source and resolved dependency edges, surfaced in the SBOM as component properties.

Features

• One command. rea <file> produces a Markdown report, an executive typeset PDF report, and a machine-readable JSON report beside the target.
• Any file. Format is detected from contents, not the file extension.
• Autonomous AI Security Assessment. When configured with an API key, REA runs an autonomous, 90-turn tool-calling analyst loop with native function calling. The model is the report's author and inspects the target itself (decompile, emulate, decrypt, audit, rules, regions, 8051, and more), writing an Executive Overview, Target Summary, Architecture Analysis, Recovered Secrets, Indicators of Compromise, and Recommendations. A grounding validator flags any quotation not backed by engine evidence.
• Deterministic Offline Fallback. With no AI API key, REA runs completely offline, generating comprehensive deterministic technical reports with zero network requests.
• Firmware & Filesystems. U-Boot legacy images (uImage), SquashFS 4.0 root filesystems, embedded Linux partitions, and raw microcontroller firmware.
• Formats. PE (EXE/DLL/SYS), ELF, Mach-O, universal binaries, firmware bundles, and raw ARM Cortex-M images.
• Static Analysis. Cryptographic hashes (MD5, SHA-1, SHA-256), import hashes (imphash), Shannon entropy, sections, memory layouts, imports, exports, and mitigations (ASLR, DEP/NX, CFG, PIE, RELRO, stack canaries).
• SBOM & Provenance. rea sbom <file> emits CycloneDX and SPDX JSON; toolchain, libraries, managed assembly references, Rust cargo-auditable crates, and build paths (PDB/DWARF) are recovered offline and tied into the evidence bundle.
• Disassembly. Recursive-descent function and call-graph recovery with inline symbol and string resolution and per-instruction semantics.
• P-Code Emulation. Sandboxed software emulator executes target code without native host execution, unmasking runtime strings, keys, and decrypted payloads.
• Protectors & Packing. High-entropy mapping, abnormal memory layouts, W^X violations, and signature matching for commercial protectors.
• Indicators. Structured extraction of URLs, IPs, emails, registry keys, file paths, commands, and user agents.
• Artifact Formats. WebAssembly, SQLite, ELF core dumps, EVM bytecode, Apple property lists, Flutter APKs, PCAP/HAR captures, JavaScript source maps, MS-DOS MZ executables, GGUF/safetensors model artifacts, and Godot PCK archives.
• Strict Safety. #![forbid(unsafe_code)] enforced across the entire codebase. Malformed or hostile inputs can never compromise the host.

<br>

Supported Targets

Format	Parsing	Analysis & Triage
PE (EXE, DLL, SYS)	Full	x86, x86-64, imports, mitigations, imphash
ELF (Linux, BSD, Android)	Full	x86/x86-64 + every SLEIGH architecture
Mach-O and universal	Full	x86/x86-64 + every SLEIGH architecture
Go Executables (ELF/PE/Mach-O)	Full	pclntab functions, modinfo dependencies, toolchain version
.NET CIL Assemblies (EXE/DLL)	Full	ECMA-335 CLI metadata, BSJB streams, CIL disassembly, token→name
Android APK & DEX Packages	In-memory	AndroidManifest.xml binary XML, permissions, classes.dex
Electron Applications (.asar)	In-memory	Archive unpacking, IPC channel extraction, hardcoded secrets
U-Boot images (uImage)	Full	Kernel identification, compression, rootfs offsets
SquashFS 4.0 Filesystems	In-memory	Unpacking, credential discovery, sub-target triage
Raw ARM Cortex-M firmware	Vector table, flash base	ARM Thumb disassembly and handler mapping
ARM Cortex-A (Linux, Android, iOS)	Full (ELF/Mach-O)	ARMv7-A and ARMv8-A/AArch64 disassembly, call graphs
WebAssembly modules (.wasm)	Full	Sections, imports, exports, function counts
SQLite databases (.db/.sqlite)	Full	Header, tables, columns, views, triggers, sampled rows
ELF core dumps	Full	Signal, registers, loadable mappings, mapped files
EVM bytecode (raw or hex)	Full	Selectors, jump targets, notable opcodes, mutability
Apple property lists & bundles	Full	Binary/XML plist, bundle identifier and version
Flutter APKs & zip archives	Full	Per-ABI native libraries, digests, Flutter assets
PCAP and HAR captures	Full	Packets, HTTP requests, hosts, HAR entries
JavaScript source maps	Full	Sources, mapping statistics, sourceMappingURL
MS-DOS MZ executables	Full	Header, relocations, 16-bit entry disassembly
Java class files (.class)	Full	Constant pool, this/super, methods, fields
Android resources.arsc	Full	Packages, string pool, configuration locales
DWARF debug info (ELF)	Full	Compile units, directories, function symbols
Windows PDB (.pdb)	Full	MSF container, info-stream version, age, GUID
YARA-subset rules	Matching	Literal/wide/nocase/hex patterns with boolean conditions
Any raw or unknown file	Identity, hashes, entropy	String extraction, indicators, entropy mapping

Non-x86 disassembly and emulation are powered by embedded SLEIGH processor specifications compiled directly by a pure-Rust engine. No Ghidra, Java, Python, or external command-line tools are required.

<br>

One-Shot Usage

The one-shot is the primary workflow: point REA at a target and read the report.

rea <path-to-target>

REA writes three files beside the target:

• re-<name>.md — GitHub-flavored Markdown report.
• re-<name>.pdf — executive typeset PDF report.
• re-<name>.json — machine-readable summary (format, architecture, hashes, mitigations, imports, capabilities, rules, vulnerabilities, crypto).

Windows x64 (PowerShell)

.\rea.exe .\sample.exe

macOS Apple Silicon

./rea ./sample

Linux x64

./rea ./sample

Focused subcommands are also available:

rea decompile <target> [symbol|0xaddr]  # recovered pseudo-C
rea strings <target>                    # extracted strings
rea decrypt <target>                    # strings hidden by single-byte ciphers
rea crypto <target>                     # cryptographic constants
rea rules <target>                      # capability rules (ATT&CK/MBC)
rea audit <target> [--sarif]            # vulnerability sinks + SARIF
rea unpack <target> [0xaddr]            # emulation-driven unpacking
rea emulate <target> [0xaddr]           # sandboxed emulation
rea dotnet <assembly>                   # .NET CIL disassembly
rea wasm <module.wasm>                  # WebAssembly sections and links
rea sqlite <database>                   # SQLite schema and rows
rea coredump <core>                     # ELF core signal, registers, mappings
rea evm <bytecode>                      # EVM selectors and mutability
rea plist <Info.plist>                  # property list and bundle metadata
rea sourcemap <app.js.map>              # source map sources and statistics
rea pcap <capture.pcap>                 # packets, HTTP requests, hosts
rea flutter <app.apk>                   # per-ABI native libraries and assets
rea dos <program.exe>                   # MZ header and 16-bit entry code
rea java <App.class>                    # Java class pool, methods, fields
rea arsc <resources.arsc>               # Android packages, strings, locales
rea dwarf <app.elf>                     # DWARF compile units and functions
rea pdb <app.pdb>                       # Windows PDB version, age, GUID
rea yara <target> <rules.yar>           # YARA-subset signature matching
rea analyze <target>                    # alias for the one-shot

<br>

MCP Server Usage (Second Option)

The MCP server exposes the same engine as tools so AI coding agents can drive the analysis interactively. It runs over stdio (JSON-RPC 2.0) via:

rea serve

One-step setup writes the agent configuration for the current binary:

rea mcp --setup    # writes mcp_config.json and opencode.jsonc for this binary
rea mcp --config   # prints ready-to-paste snippets for every supported agent

Claude Code

claude mcp add rea rea serve

Then ask, for example:

Open ./sample.exe with rea, decompile the entry point, run the capability rules,
and explain what it does.

OpenCode

Add REA to opencode.json (project) or ~/.config/opencode/opencode.jsonc (global):

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "rea": {
      "type": "local",
      "command": ["/absolute/path/to/rea", "serve"],
      "enabled": true
    }
  }
}

rea mcp --setup writes this for you with the correct absolute path. Restart OpenCode, then ask:

Use rea to open ./sample.exe, list the key functions, and decompile the entry point.

Per-Platform MCP Setup

Windows x64 (PowerShell)

.\rea.exe mcp --setup
# Claude Code:
claude mcp add rea "$PWD\rea.exe" serve

macOS Apple Silicon

./rea mcp --setup
# Claude Code:
claude mcp add rea "$PWD/rea" serve

Linux x64

./rea mcp --setup
# Claude Code:
claude mcp add rea "$PWD/rea" serve

The server exposes 31 tools: open_binary, binary_session, list_functions, decompile_function, find_xrefs, search_strings, analyze_firmware, analyze_javascript, generate_dossier, inspect_go, inspect_dotnet, inspect_android, run_rules, audit, inspect_crypto, decrypt_strings, emulate, unpack, inspect_wasm, inspect_sqlite, inspect_coredump, inspect_evm, inspect_plist, inspect_sourcemap, inspect_capture, inspect_flutter, inspect_dos, inspect_dwarf, inspect_java, inspect_arsc, inspect_pdb.

<br>

AI Analyst Configuration

To enable the autonomous tool-calling analyst loop, provide an OpenRouter API key. REA locates the key from environment variables or a .env file in the current or ancestor directories:

# In .env or shell environment:
OPENROUTER_API_KEY="sk-or-v1-..."

# Optional model selection (defaults to high-throughput free tier):
OPENROUTER_MODEL="nvidia/nemotron-3.5-lightning:free"

If no key is configured, REA operates in purely deterministic offline mode.

<br>

Build from Source

Install the stable Rust toolchain, clone the repository, and build on your operating system:

git clone https://github.com/mytechnotalent/rea.git
cd rea

Windows x64 (PowerShell)

rustup target add x86_64-pc-windows-msvc
cargo build --release -p rea --target x86_64-pc-windows-msvc

Output: target\x86_64-pc-windows-msvc\release\rea.exe

macOS Apple Silicon

rustup target add aarch64-apple-darwin
cargo build --release -p rea --target aarch64-apple-darwin

Output: target/aarch64-apple-darwin/release/rea

Linux x64

rustup target add x86_64-unknown-linux-gnu
cargo build --release -p rea --target x86_64-unknown-linux-gnu

Output: target/x86_64-unknown-linux-gnu/release/rea

Install a native build on your PATH with:

cargo install --path crates/rea

Prebuilt binaries for macOS Apple Silicon (aarch64-apple-darwin) are located in releases/. To build for Windows or Linux, follow the commands above.

<br>

Quality & Verification

cargo xtask check

Enforces strict zero-violation coding standards (CONVENTIONS.md), runs cargo fmt --check, denies all Clippy warnings (cargo clippy --workspace --all-targets -- -D warnings), and executes the complete workspace test suite.

<br>

License

MIT License