AutoRE-CLI
Bounded static reverse engineering for human analysts and AI agents.
AutoRE-CLI turns ELF, PE/COFF, Mach-O, object files, and explicitly identified raw shellcode into traceable JSON, CFG/IL, readable pseudo output, language evidence, and safe next actions—without executing target bytes.

Download · Agent Quick Start · CLI Quick Start · Why AutoRE-CLI · Security · FAQ
Why AutoRE-CLI
- Evidence before confidence: findings remain explicitly
validated,inferred,unresolved, ornot_claimed. - Agent-sized context: bounded JSON, budgets, warnings, stop reasons,
verified bundles, and exact
next_actions[].argvprevent output floods. - Static-only boundary: no target, recovered payload, shellcode, embedded object, or target-derived artifact is executed.
- Layered inspection: disassembly, CFG, LLIL, MLIL, HLIL, CUSTOM IL, pseudo output, types, calls, data references, PE inventories, and Go/Rust evidence.
- Repeatable investigations: archive, replay, semantic diff, and batch comparison preserve the evidence path.
- Offline and traceable: installation performs no network access; every release binary has an exact target, byte count, source revision, signing disposition, and SHA-256.
Where It Fits
| Tool | Best fit | AutoRE-CLI difference |
|---|---|---|
| Ghidra, Rizin, angr | Extensible interactive frameworks and custom analysis | AutoRE-CLI is a prebuilt, CLI-first bounded evidence surface for automation and agents |
| Ghidra MCP integrations | Conversational control of an existing Ghidra environment | AutoRE-CLI needs no GUI or analysis server and keeps static-only limits in the output contract |
| capa | Rule-based capability identification | AutoRE-CLI exposes functions, CFG/IL, pseudo output, references, language evidence, and comparison workflows |
| General binary-analysis Skills | Methodology layered over separately installed tools | The included Skill and CLI share one verified action, artifact, and evidence contract |
AutoRE-CLI is not a debugger, emulator, sandbox, or claim of perfect source recovery. Use the larger frameworks when you need plugins, interactive GUI workflows, symbolic execution, or dynamic observation.
Install
Requirements: Python 3.9 or newer and a supported host. No Rust toolchain or source checkout is required.
Platform Release
Download the archive for your host from GitHub Releases, then verify and install it:
tar -xzf AutoRE-CLI-0.1.1-macos-arm64.tar.gz
cd AutoRE-CLI-0.1.1-macos-arm64
./verify.sh
./install.sh
auto-re-cli --version
Replace macos-arm64 with macos-x86_64, linux-x86_64, or linux-arm64.
Windows users should extract AutoRE-CLI-0.1.1-windows-x86_64.zip and run:
py -3 scripts/autore_distribution.py verify
py -3 scripts/autore_distribution.py install
auto-re-cli --version
Each approximately 7–8 MB platform archive contains one binary, the Agent Skill, offline installers, provenance, checksums, and notices. You can also clone this repository to obtain the complete platform matrix:
git clone --depth 1 https://github.com/timwhitez/AutoRE-CLI.git
cd AutoRE-CLI
./verify.sh
./install.sh
Package Managers
Package managers install the CLI only. Install the Agent Skill separately, or use the platform archive above for the complete verified distribution.
# Agent Skill for Codex, Claude Code, Cursor, and other supported clients
npx skills add \
https://github.com/timwhitez/AutoRE-CLI/releases/download/v0.1.1/AutoRE-CLI-0.1.1-auto-re-skill.zip -g
# Scoop CLI on Windows
scoop install https://raw.githubusercontent.com/timwhitez/AutoRE-CLI/main/packaging/scoop/autore-cli.json
The repository includes the versioned Homebrew formula and Scoop manifest. The Homebrew formula is ready for a dedicated Tap; until that Tap is published, use the verified macOS or Linux platform archive. Package-manager availability can lag behind a GitHub Release.
The default installation copies:
auto-re-cli(orauto-re-cli.exe) to${AUTORE_INSTALL_DIR:-$HOME/.local/bin};- the Skill to
${TRAE_HOME:-$HOME/.trae}/skills/auto-re; - the Skill to
$HOME/.agents/skills/auto-re.
The installer verifies the complete extracted distribution, probes only the
verified binary with --version, copies rather than symlinks, records managed
markers, and refuses to overwrite unmanaged destinations unless
--replace-unmanaged is explicit.
Common options:
./install.sh --dry-run
./install.sh --cli-only
./install.sh --skill-only --agents trae
./install.sh --skill-only --agents codex
./install.sh --skill-only --agents both
./install.sh --install-dir "$HOME/bin"
./install.sh --skill-only --agents codex --codex-home "$HOME/.codex"
--codex-home selects the legacy <home>/skills compatibility location. The
current default is $HOME/.agents/skills.
Update And Uninstall
Extract a newer release, then rerun ./verify.sh and ./install.sh. Managed
drift, same-version binary repacks, downgrades, and extra Skill files fail
closed.
./uninstall.sh
./uninstall.sh --force-managed
--force-managed applies only to marker-listed files whose content changed.
It never authorizes removal of unrelated parent directories or extra user
files.
Agent Quick Start
Restart or refresh your agent after installation, then invoke:
Use $auto-re to statically inspect ./sample.exe and write bounded,
evidence-backed findings under ./analysis-results. Never execute the sample or
any target-derived artifact.
The Skill verifies the CLI, validates bundle ownership/size/SHA-256 before reading payloads, separates evidence strength, follows one relevant static continuation at a time, and stops at explicit evidence or budget boundaries.
To prepare one emitted action without shell interpolation:
python3 skills/auto-re/scripts/run_next_action.py \
./analysis-results/sample.bundle/manifest.json \
--action-stage function.selected \
--output ./analysis-results/function-selected.json \
--dry-run
Review the exact argument vector before removing --dry-run. Actions that
already own a bundle/spill sink reject an additional --output.
CLI Quick Start
Bounded Context Bundle
mkdir -p ./analysis-results
auto-re-cli report ./sample.exe \
--format json \
--json-profile ai \
--sections binary,summary,inspections,flow,functions,types \
--limit 8 \
--bundle-dir ./analysis-results/sample.bundle \
--output ./analysis-results/sample.bundle/manifest.json
python3 skills/auto-re/scripts/verify_bundle.py \
./analysis-results/sample.bundle/manifest.json
The manifest records owned payload paths, byte counts, SHA-256, warnings, budgets, completion state, current findings, and bounded next actions.
Focused Function
auto-re-cli function ./sample.exe \
--addr 0x401000 \
--format json \
--output ./analysis-results/function-401000.json
auto-re-cli dump-il ./sample.exe \
--level custom \
--addr 0x401000 \
--format json \
--output ./analysis-results/custom-il-401000.json
Reproduce The Safe Demo
examples/controlled/fixture.c is a benign,
independently authored fixture. Build it as a non-runnable object and follow
the commands in
examples/controlled/README.md. The checked
demo found one AArch64 function, completed without truncation or warnings, and
kept absent type evidence as not_claimed.
Explicit Raw Shellcode
auto-re-cli report ./stage.bin \
--raw-shellcode \
--arch x86 \
--base-address 0x1000 \
--entry-address 0x1000 \
--format json \
--json-profile ai \
--sections summary,flow,functions \
--bundle-dir ./analysis-results/raw.bundle \
--output ./analysis-results/raw.bundle/manifest.json
Raw architecture, base, and entry are caller-provided facts. Never guess them from a filename.
Analysis Surfaces
| Task | Command |
|---|---|
| Aggregate triage | report, analyze, decompile |
| Function and local slices | function, slice-function |
| Intermediate language and CFG | dump-il, dump-cfg, inspect-passes |
| Static relationships | inspect-flow, call-graph, data-xrefs, aarch64-refs |
| PE inventories | pe-resources, pe-strings |
| Language evidence | inspect-go, inspect-rust, inspect-types |
| Protection evidence | inspect-die, inspect-upx, inspect-vmp |
| Repeatable comparison | batch, archive, replay, diff, batch-diff |
Run auto-re-cli <command> --help for the exact options supported by the
installed version.
Platforms
| Host | Release target | Status |
|---|---|---|
| macOS Apple Silicon | macos-arm64 |
Published; ad-hoc signed |
| macOS Intel | macos-x86_64 |
Published; ad-hoc signed |
| Linux x86-64 | linux-x86_64 |
Published |
| Linux AArch64 | linux-arm64 |
Published |
| Windows x86-64 | windows-x86_64 |
Published; unsigned |
The macOS binaries are not Developer ID signed or notarized. The Windows binary supports Windows 10 or newer and is not Authenticode signed.
Integrity And Provenance
Run ./verify.sh before installation. It validates:
- every
SHA256SUMSrow and exact public file set; - every binary target/path/size/hash in
manifest/release.json; - the exact managed Skill file set;
- the MIT-only project license boundary;
- the personal publisher and repository identity;
- absence of source roots, Rust source, private paths, samples, internal artifacts, and governed internal-account markers.
manifest/release.json records the publisher, repository URL, release version,
distribution scope, source revision, toolchain, target matrix, and static
safety flags. THIRD_PARTY_LICENSES.md and licenses/third-party/ preserve
dependency notices separately from the project license.
Public Repository Boundary
This repository is the MIT-licensed public binary distribution of Auto-RE. It contains compiled binaries, the open Agent Skill and helpers, installers, verifier, release automation, controlled demo source, checksums, provenance, notices, and documentation.
The engine's Rust implementation, private specifications, samples, and analysis output are not published here. “Open source” in this repository applies to the public scripts, Skill, examples, automation, and documentation; it does not imply that the binary engine implementation is available.
Repository Layout
assets/ public demo and social artwork
bin/ platform binaries
examples/controlled/ benign reproducible demo source
skills/auto-re/ Agent Skill and deterministic helpers
scripts/autore_distribution.py
scripts/build_release_assets.py
manifest/release.json machine-readable provenance
licenses/third-party/ dependency notices
SHA256SUMS outer integrity manifest
install.sh / uninstall.sh managed local installation
verify.sh fail-closed distribution verifier
Support
- Bugs and documentation: issue tracker
- Use cases and integration requests: public discussion
- Security: private vulnerability report
- Contribution boundary: CONTRIBUTING.md
- Release history: CHANGELOG.md
- Maintainer: @timwhitez
Do not upload malware samples, recovered payloads, secrets, private paths, or proprietary analysis output.
License
The public AutoRE-CLI distribution is licensed under the MIT License. Third-party dependencies retain their own licenses, including Apache-licensed dependencies where applicable; see THIRD_PARTY_LICENSES.md.
No comments yet
Be the first to share your take.