RAPTOR

No entries match these filters. Clear filters

1. docs update Oct 8, 2026 · issue 097

   RAPTOR is an agentic CLI for end-to-end vulnerability research: scan, multi-stage LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. exploit validation (Stages A-F), SCASCASoftware composition analysis, identifying third-party dependencies and checking them for known vulnerabilities and license issues. It covers software a team uses rather than writes itself. with CycloneDXCycloneDXAn open standard for describing software components and their dependencies, vulnerabilities and other supply-chain information. Tools use it to exchange bills of materials and related evidence. SBOMSBOMA software bill of materials, an inventory of the components and versions inside an application. It helps teams find affected software when a dependency has a vulnerability or licensing issue. output, binary analysis, and air-gapped pipeline support.

   • Adds --max-cost-usd flag to cap per-run LLM analysis spend (default $10) when running the full validation pipeline.
   • Adds --codeql flag to raptor to include CodeQL alongside Semgrep and Coccinelle in a scan.
   • Adds --threat-model, --threat-model-refresh, and --threat-model-use-stale flags to control automated threat-model generation and freshness enforcement, writing threat-model.json and THREAT_MODEL.md per project.
   • Adds --project and --continue flags to the raptor launcher for named workspace management and session resumption.
   • Adds --model flag to the raptor launcher to select the orchestration LLM at invocation time.

   +14 more show less

   • Adds /agentic command for a full autonomous scan-validate-exploit-patch workflow.
   • Adds /scan command for static analysis using Semgrep and CodeQL with deduplication and SARIFSARIFStatic Analysis Results Interchange Format, a standard for sharing code-analysis findings, including locations and rule details. It lets scanners report results to a common review interface. output.
   • Adds /understand command to map attack surfaceattack surfaceThe exposed systems, interfaces and access points an attacker could try to use. It includes internet-facing services as well as identities, software and other reachable resources., trace data flows, and hunt vulnerability variants before scanning.
   • Adds /binary command for black-box binary investigation with subcommands graph, trace-parser, harness, and investigate --active.
   • Adds /sca command for software composition analysissoftware composition analysisSoftware composition analysis, identifying third-party dependencies and checking them for known vulnerabilities and license issues. It covers software a team uses rather than writes itself. producing sbom.cdx.json (CycloneDX with VEXVEXA machine-readable statement from a vendor saying whether a CVE in a component actually affects their product. It is how a scanner is told to stop flagging a vulnerability that is not reachable.), findings.sarif, and report.md, with subcommands fix, check, upgrade, diff, verify, health, render, suppress, and clean-cache.
   • Adds /cve-env command to build and verify a Docker environment running a CVE's affected application at its pre-patch version.
   • Adds /exploit command to generate proof-of-concept exploit code for confirmed vulnerabilities.
   • Adds /patch command to generate secure patches for confirmed vulnerabilities.
   • Adds /crash-analysis command for autonomous root-cause analysis of C/C++ crashes.
   • New integrationAdds /oss-forensics command for evidence-backed forensic investigation of GitHub repositories.
   • Adds multi-stage exploitability validation pipeline (Stages A-F) covering pattern validation, attacker preconditions, reachability, final triage, binary feasibility, and self-review.
   • Adds optional Z3 SMT integration (pip install z3-solver) for dataflow pre-screening in CodeQL and one-gadget constraint analysis in binary feasibility.
   • Sandboxes untrusted-content subprocesses using Linux namespaces, Landlock, and seccomp (see docs/sandbox.md).
   • Strips environment variables that could inject code into the launcher chain at startup via core/security/_dangerous_env_strip.sh.

   19 details in Everything

   • New detection Custom Semgrep rules for Python, Go, Java, JS/TS
   • New detection Custom Coccinelle rules for C/C++ memory safety and integer bugs
   • New integration Semgrep registry pack offline cache tool
   • New integration Ghidra RE bridge integration
   • New integration Frida dynamic instrumentation integration
   • New integration ffuf and nuclei integration for web scanning
   • New integration AFL++ binary fuzzing integration
   • New supported platform VS Code devcontainer support
2. docs update Sep 24, 2026 · issue 084

   RAPTOR adds /openant LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. source-code scan command and fully standalone (no Claude Code) execution path⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   └──▷ TRY IT
   Run a full scan-to-analysis pipeline on an airgapped box with no Claude Code or Anthropic connectivity.
   copy

   $ export _RAPTOR_TRUSTED=1
   export OLLAMA_HOST=http://localhost:11434
   python3 raptor.py agentic --repo /path/to/code

   Use raptor-project-manager to create and activate a project, then run a standalone scan that picks up the project context automatically.
   copy

   $ export _RAPTOR_TRUSTED=1
   libexec/raptor-project-manager create myapp --target /path/to/code
   libexec/raptor-project-manager use myapp
   python3 raptor.py agentic --repo /path/to/code
   libexec/raptor-project-manager findings

   • Adds /openant slash command for OpenAnt LLM source-code scanning: ASTASTAn abstract syntax tree, a structured representation of parsed code. Tools inspect or transform this tree to understand expressions and statements rather than just matching text. analysis plus per-function LLM reasoning (Beta).⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Adds a fully standalone execution path via python3 raptor.py with no Claude Code dependency, supporting doctor, agentic --repo, and scan --repo subcommands.
   • Adds _RAPTOR_TRUSTED=1 environment variable to allow libexec/raptor-* scripts (including raptor-project-manager) to run outside a Claude Code session.
   • Adds raptor-project-manager subcommands create, use, status, and findings for project lifecycle management in standalone mode.
   • Supports Ollama via OLLAMA_HOST and models.json for fully local, airgapped scan/analysis/exploit pipelines without Anthropic connectivity.

   5 details in Everything

3. docs update Sep 24, 2026 · issue 084

   RAPTOR: Adds docs/troubleshooting.md covering self-test, sandbox setup errors on Ubuntu 24.04+, and EDREDREndpoint detection and response, monitoring computers for suspicious behavior and providing investigation and response controls, such as isolating a device. interaction.

   └──▷ TRY IT
   Run a full scan-to-analysis pipeline on an airgapped host or without a Claude Code subscription.
   copy

   $ python3 raptor.py agentic --repo /path/to/code

   Set up a named project and run analysis against it in standalone mode — useful for recurring scans across multiple targets.
   copy

   $ export _RAPTOR_TRUSTED=1
   libexec/raptor-project-manager create myapp --target /path/to/code
   libexec/raptor-project-manager use myapp
   python3 raptor.py agentic --repo /path/to/code

   • Adds docs/agent-security.md covering agent capabilities, tool boundaries, network controls, and human approval workflows.

   1 detail in Everything

4. docs update Sep 15, 2026 · issue 075

   RAPTOR debuts with 30+ slash commands covering autonomous vuln validation, binary analysis, SCASCASoftware composition analysis, identifying third-party dependencies and checking them for known vulnerabilities and license issues. It covers software a team uses rather than writes itself., fuzzing, and web scanning inside Claude Code sessions.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   └──▷ TRY IT
   Run a full autonomous scan-validate-exploit-patch pipeline against a local repo while capping LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. API spend at $20 and anchoring analysis to a generated threat model.
   copy

   $ raptor /path/to/target --project myapp --threat-model --max-cost-usd 20

   Mount a target repository into the RAPTOR devcontainer for isolated analysis without installing dependencies locally.
   copy

   $ docker build -f .devcontainer/Dockerfile -t raptor:latest . && docker run --rm -it -v "$(pwd):/workspaces/raptor" -v "/path/to/target-folder:/workspaces/target" raptor:latest raptor

   • Launches the raptor binary (from bin/raptor) with flags --project, --continue, and --model to start a session with pre-flight checks, coverage tracking, and environment sanitisation.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Adds --max-cost-usd flag to cap per-run LLM API spend (default $10) across the analysis dispatch layer.
   • Adds --codeql flag to /scan to include CodeQL deep analysis alongside Semgrep and Coccinelle.
   • Adds --threat-model and --threat-model-refresh flags to /agentic to generate or refresh threat-model.json and THREAT_MODEL.md before analysis; --threat-model-use-stale allows stale fallback maps.
   • Adds /agentic command for a full autonomous workflow: scan, validate, exploit, and patch in a single pass.

   +32 more show less

   • Adds /scan command for static analysis with Semgrep and CodeQL, producing SARIFSARIFStatic Analysis Results Interchange Format, a standard for sharing code-analysis findings, including locations and rule details. It lets scanners report results to a common review interface. output.
   • Adds /validate command running a multi-stage exploitability validation pipeline (Stages A-F) against existing findings.
   • Adds /binary command for black-box binary investigation with subcommands graph --edges --json, trace-parser, investigate --active, and harness for runtime evidence and graph queries.
   • Adds /ghidra command to bridge Ghidra RE: attach or import .gpr projects, run cross-version diffs, and export findings.
   • Adds /sca command for software composition analysissoftware composition analysisSoftware composition analysis, identifying third-party dependencies and checking them for known vulnerabilities and license issues. It covers software a team uses rather than writes itself. covering manifests, lockfiles, workflow dependencies, and container sources; outputs findings.json, report.md, sbom.cdx.json (CycloneDXCycloneDXAn open standard for describing software components and their dependencies, vulnerabilities and other supply-chain information. Tools use it to exchange bills of materials and related evidence. with VEXVEXA machine-readable statement from a vendor saying whether a CVE in a component actually affects their product. It is how a scanner is told to stop flagging a vulnerability that is not reachable.), and findings.sarif; subcommands include fix, check, upgrade, diff, verify, health, render, suppress, and clean-cache.
   • Adds /cve-diff command to discover and diff the fix commit for a CVE across OSVOSVOpen Source Vulnerabilities, a vulnerability database and schema maintained by Google that gives cyber tools a machine-readable, cross-ecosystem format for querying and sharing CVE-style advisories., NVDNVDThe National Vulnerability Database, maintained by NIST, is the U.S. government repository of CVE records enriched with CVSS scores and metadata; cyber tools query it to drive vulnerability prioritization., GitHub, and GitLab.
   • Adds /cve-env command to build and verify a Docker environment running a CVE's affected application at its pre-patch version.
   • Adds /exploit command to generate proof-of-concept exploit code for confirmed vulnerabilities.
   • Adds /patch command to generate secure patches for confirmed vulnerabilities.
   • Adds /fuzz command for binary fuzzing with AFL++ and crash analysis.
   • Adds /crash-analysis command for autonomous root-cause analysis of C/C++ crashes.
   • Adds /web command for web application scanning: crawl, ffuf/nuclei integration, oracle-verified injection, and blind SSRFSSRFServer-side request forgery, an attack that tricks a server into making requests chosen by an attacker. Those requests can reach internal services the attacker cannot access directly. callbacks.
   • Adds /understand command to map attack surfaceattack surfaceThe exposed systems, interfaces and access points an attacker could try to use. It includes internet-facing services as well as identities, software and other reachable resources., trace data flows, and hunt vulnerability variants.
   • Adds /oss-forensics command for evidence-backed forensic investigation of GitHub repositories.
   • Adds /threat-model command to create, inspect, and maintain per-project threat models.
   • Adds /sage command as a persistent memory layer with store, recall, link, and corroborate operations.
   • Adds /frida command for dynamic instrumentation via Frida.
   • Adds /codeql command for CodeQL-only deep analysis with Z3 SMT dataflow pre-screening.
   • Adds /analyze command to run LLM analysis against existing SARIF findings without re-scanning.
   • Adds /diagram command to generate Mermaid visual maps from /understand and /validate JSON outputs.
   • Adds /audit command for hypothesis-driven, tool-grounded systematic code review.
   • Adds /review command to query audit state: findings, gaps, coverage, and operator notes.
   • Adds /annotate command to attach free-form per-function prose annotations as operator review notes.
   • Adds /project command to manage named workspaces that organise runs and track findings over time.
   • Adds /describe command to profile a target: language mix, build system, tool gaps, and cost estimate (read-only).
   • Adds /scorecard command to inspect per-model reliability across decision classes.
   • Adds /ask command to send a free-form prompt to any configured LLM model.
   • Supports optional Z3 SMT integration (pip install z3-solver) for CodeQL dataflow pre-screening and binary one-gadget constraint analysis.
   • Supports offline and air-gapped pipelines via engine/semgrep/rules/ local rules and engine/semgrep/tools/cache-packs.py to pre-cache registry packs (p/security-audit, p/owasp-top-ten, etc.) as a zip bundle for transfer.
   • Sandboxes untrusted-content subprocesses using Linux namespaces, Landlock, and seccomp, stripping dangerous environment variables via core/security/_dangerous_env_strip.sh.
   • Supports devcontainer deployment via .devcontainer/Dockerfile and .devcontainer/devcontainer.json with bind-mount configuration for target folders.
   • Supports multi-model analysis dispatch (Claude, OpenAI, GPT, Gemini) configurable per run when API keys for additional providers are supplied.

   37 details in Everything

5. docs update Sep 15, 2026 · issue 075

   RAPTOR: Adds subprocess sandboxingsandboxingRunning code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls. using Linux namespaces, Landlock, and seccomp for processes handling untrusted content, with the threat model documented in docs/sandbox.md.

   └──▷ TRY IT
   Run a full autonomous scan-validate-exploit-patch cycle against a codebase with a $15 cost ceiling, using a named project workspace to track findings over time.
   copy

   $ raptor /path/to/target --project my-audit
   # Then inside the session:
   /agentic --codeql --threat-model --max-cost-usd 15

   Mount a target repository into the RAPTOR devcontainer and attach the rr deterministic debugger for crash-analysis work.
   copy

   $ docker run --privileged \
     -v "$(pwd):/workspaces/raptor" \
     -v "/path/to/target-folder:/workspaces/target" \
     raptor:latest

   • Supports devcontainer deployment with target-folder mounts via the mounts section of .devcontainer/devcontainer.json, and --privileged for rr deterministic debugger access.

   1 detail in Everything

6. v3.1.0 Sep 7, 2026 · issue 068

   RAPTOR v3.1.0 lands bulk audit mode, OCIOCIOpen Container Initiative standard, maintained by the Linux Foundation, defining the image format and runtime spec that container tools share so images built once run anywhere OCI-compliant./Dockerfile substrate, SCASCASoftware composition analysis, identifying third-party dependencies and checking them for known vulnerabilities and license issues. It covers software a team uses rather than writes itself. supply-chain detection, multi-LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. provider support, and FridaFridaA dynamic instrumentation toolkit that lets developers inject JavaScript or native code into running processes on Windows, macOS, Linux, iOS, and Android to inspect and modify behavior at runtime. reachability integration.

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v3.1.0 https://github.com/gadievron/raptor.git
   # already have the repo? check out this version:
   $ git checkout v3.1.0

   └──▷ TRY IT
   Audit a target's sandbox behavior before enforcing — log every syscall that would have been blocked without actually blocking anything, with a cap to avoid log flooding.
   copy

   $ /scan /path/to/target --audit --audit-verbose --audit-budget 500

   Run the compiler-analyzer channel on a C/C++ codebase alongside Semgrep, capping the number of translation units to keep the scan tractable on a large repo.
   copy

   $ /scan /path/to/c-project --compiler-scan --compiler-scan-max-tus 500 --build-command 'make -j4'

   Scan a macro-heavy C/C++ codebase and catch bugs that only appear in expanded preprocessor output, which plain Semgrep misses.
   copy

   $ /scan /path/to/c-project --expanded-semgrep --languages c,cpp

   • Adds --audit-verbose flag for strace-style verbose output during sandbox audit runs.
   • Adds --audit-budget <n> flag to cap the maximum number of audit events collected per run.
   • Adds --compiler-scan / --no-compiler-scan flags to enable or disable a compiler-analyzer channel using gcc -fanalyzer or clang --analyze.
   • Adds --compiler-scan-max-tus <n> flag to cap the number of translation units analyzed by --compiler-scan (default 2000).
   • Adds --expanded-semgrep flag to re-run Semgrep rules over preprocessor-expanded views of macro-heavy C/C++ translation units.

   +14 more show less

   • Adds OCI registry, Dockerfile, and tar as scan substrates, enabling RAPTOR to analyze container images and tarballs directly.
   • Adds a full SCA (software composition analysissoftware composition analysisSoftware composition analysis, identifying third-party dependencies and checking them for known vulnerabilities and license issues. It covers software a team uses rather than writes itself.) mechanical pipeline with supply-chain detector and first-party hardening.
   • Adds cve-env as an integrated package (packages/cve_env) for CVE environment modeling (Phase 1).
   • Adds full multi-LLM provider support for the analysis dispatch layer, with Mantle and InvokeModel coexistence.
   • Adds source intelligence substrate with a minimal axis-1 rule for source-level intel gathering.
   • Adds a corpus runner with rule evaluation and label linting for calibrating detection rules.
   • Adds per-function annotations emitted across validation stages A, B, and D via the validate module.
   • Adds mechanical sink discovery and framework API detection (MAP-5f) to the understand module.
   • Adds agentic checker synthesis — the system now generates a checker per confirmed finding.
   • Adds JS and Go function-as-argument framework registration for reachability analysis.
   • Adds a persistent verdict-bearing coverage store wired into the live pipeline.
   • Adds Frida evidence pipeline with full consumer wiring and reachability integration.
   • Adds a web correlation tier with expectations, check legs, and a poisoned-render oracle for web vulnerability analysis.
   • Adds proxy-hosts auto-calibration for CodeQL analysis.
   • Adds --audit-verbose flag for strace-style verbose output during sandbox audit runs.
   • Adds --audit-budget <n> flag to cap the maximum number of audit events collected per run.
   • Adds --compiler-scan / --no-compiler-scan flags to enable or disable a compiler-analyzer channel using gcc -fanalyzer or clang --analyze.
   • Adds --compiler-scan-max-tus <n> flag to cap the number of translation units analyzed by --compiler-scan (default 2000).
   • Adds --expanded-semgrep flag to re-run Semgrep rules over preprocessor-expanded views of macro-heavy C/C++ translation units.

   +12 more show less

   • Adds OCI registry, Dockerfile, and tar as scan substrates, enabling RAPTOR to analyze container images and tarballs directly.
   • Adds a full SCA (software composition analysissoftware composition analysisSoftware composition analysis, identifying third-party dependencies and checking them for known vulnerabilities and license issues. It covers software a team uses rather than writes itself.) mechanical pipeline with supply-chain detector and first-party hardening.
   • Adds cve-env as an integrated package (packages/cve_env) for CVE environment modeling (Phase 1).
   • Adds full multi-LLM provider support for the analysis dispatch layer, with Mantle and InvokeModel coexistence.
   • Adds source intelligence substrate with a minimal axis-1 rule for source-level intel gathering.
   • Adds a corpus runner with rule evaluation and label linting for calibrating detection rules.
   • Adds per-function annotations emitted across validation stages A, B, and D via the validate module.
   • Adds mechanical sink discovery and framework API detection (MAP-5f) to the understand module.
   • Adds agentic checker synthesis — the system now generates a checker per confirmed finding.
   • Adds a persistent verdict-bearing coverage store wired into the live pipeline.
   • Adds Frida evidence pipeline with full consumer wiring and reachability integration.
   • Adds a web correlation tier with expectations, check legs, and a poisoned-render oracle for web vulnerability analysis.

   2 more in Everything

7. v3.1.0 Sep 7, 2026 · issue 011

   RAPTOR v3.1.0 lands bulk audit mode, OCIOCIOpen Container Initiative standard, maintained by the Linux Foundation, defining the image format and runtime spec that container tools share so images built once run anywhere OCI-compliant./Dockerfile substrate, SCASCASoftware composition analysis, identifying third-party dependencies and checking them for known vulnerabilities and license issues. It covers software a team uses rather than writes itself. supply-chain detection, multi-LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. provider support, and FridaFridaA dynamic instrumentation toolkit that lets developers inject JavaScript or native code into running processes on Windows, macOS, Linux, iOS, and Android to inspect and modify behavior at runtime. reachability integration.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v3.1.0 https://github.com/gadievron/raptor.git
   # already have the repo? check out this version:
   $ git checkout v3.1.0

   └──▷ TRY IT
   Audit a target's sandbox behavior before enforcing — log every syscall that would have been blocked without actually blocking anything, with a cap to avoid log flooding.
   copy

   $ /scan /path/to/target --audit --audit-verbose --audit-budget 500

   Run the compiler-analyzer channel on a C/C++ codebase alongside Semgrep, capping the number of translation units to keep the scan tractable on a large repo.
   copy

   $ /scan /path/to/c-project --compiler-scan --compiler-scan-max-tus 500 --build-command 'make -j4'

   Scan a macro-heavy C/C++ codebase and catch bugs that only appear in expanded preprocessor output, which plain Semgrep misses.
   copy

   $ /scan /path/to/c-project --expanded-semgrep --languages c,cpp

   • Adds --audit-verbose flag for strace-style verbose output during sandbox audit runs.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Adds --audit-budget <n> flag to cap the maximum number of audit events collected per run.
   • Adds --compiler-scan / --no-compiler-scan flags to enable or disable a compiler-analyzer channel using gcc -fanalyzer or clang --analyze.
   • Adds --compiler-scan-max-tus <n> flag to cap the number of translation units analyzed by --compiler-scan (default 2000).
   • Adds --expanded-semgrep flag to re-run Semgrep rules over preprocessor-expanded views of macro-heavy C/C++ translation units.

   +14 more show less

   • Adds OCI registry, Dockerfile, and tar as scan substrates, enabling RAPTOR to analyze container images and tarballs directly.
   • Adds a full SCA (software composition analysissoftware composition analysisSoftware composition analysis, identifying third-party dependencies and checking them for known vulnerabilities and license issues. It covers software a team uses rather than writes itself.) mechanical pipeline with supply-chain detector and first-party hardening.
   • Adds cve-env as an integrated package (packages/cve_env) for CVE environment modeling (Phase 1).
   • Adds full multi-LLM provider support for the analysis dispatch layer, with Mantle and InvokeModel coexistence.
   • Adds source intelligence substrate with a minimal axis-1 rule for source-level intel gathering.
   • Adds a corpus runner with rule evaluation and label linting for calibrating detection rules.
   • Adds per-function annotations emitted across validation stages A, B, and D via the validate module.
   • Adds mechanical sink discovery and framework API detection (MAP-5f) to the understand module.
   • Adds agentic checker synthesis — the system now generates a checker per confirmed finding.
   • Adds JS and Go function-as-argument framework registration for reachability analysis.
   • Adds a persistent verdict-bearing coverage store wired into the live pipeline.
   • Adds Frida evidence pipeline with full consumer wiring and reachability integration.
   • Adds a web correlation tier with expectations, check legs, and a poisoned-render oracle for web vulnerability analysis.
   • Adds proxy-hosts auto-calibration for CodeQL analysis.
   • Adds --audit-verbose flag for strace-style verbose output during sandbox audit runs.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Adds --audit-budget <n> flag to cap the maximum number of audit events collected per run.
   • Adds --compiler-scan / --no-compiler-scan flags to enable or disable a compiler-analyzer channel using gcc -fanalyzer or clang --analyze.
   • Adds --compiler-scan-max-tus <n> flag to cap the number of translation units analyzed by --compiler-scan (default 2000).
   • Adds --expanded-semgrep flag to re-run Semgrep rules over preprocessor-expanded views of macro-heavy C/C++ translation units.

   +12 more show less

   • Adds OCI registry, Dockerfile, and tar as scan substrates, enabling RAPTOR to analyze container images and tarballs directly.
   • Adds a full SCA (software composition analysissoftware composition analysisSoftware composition analysis, identifying third-party dependencies and checking them for known vulnerabilities and license issues. It covers software a team uses rather than writes itself.) mechanical pipeline with supply-chain detector and first-party hardening.
   • Adds cve-env as an integrated package (packages/cve_env) for CVE environment modeling (Phase 1).
   • Adds full multi-LLM provider support for the analysis dispatch layer, with Mantle and InvokeModel coexistence.
   • Adds source intelligence substrate with a minimal axis-1 rule for source-level intel gathering.
   • Adds a corpus runner with rule evaluation and label linting for calibrating detection rules.
   • Adds per-function annotations emitted across validation stages A, B, and D via the validate module.
   • Adds mechanical sink discovery and framework API detection (MAP-5f) to the understand module.
   • Adds agentic checker synthesis — the system now generates a checker per confirmed finding.
   • Adds a persistent verdict-bearing coverage store wired into the live pipeline.
   • Adds Frida evidence pipeline with full consumer wiring and reachability integration.
   • Adds a web correlation tier with expectations, check legs, and a poisoned-render oracle for web vulnerability analysis.

   2 more in Everything

8. v3.0.0 Apr 23, 2026 · release notes

   RAPTOR v3.0.0 introduces /agentic, /codeql, and Z3 SMT constraint solving for static analysis.

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v3.0.0 https://github.com/gadievron/raptor.git
   # already have the repo? check out this version:
   $ git checkout v3.0.0

   └──▷ TRY IT
   Run a standalone CodeQL analysis with Z3 SMT constraint solving for deep dataflow tracing.
   copy

   $ python3 raptor.py /codeql <path>

   • Adds python3 raptor.py <command> as the Python CLI entry point for CI/CD integration
   • Adds /codeql subcommand for CodeQL-only deep analysis with Z3 SMT dataflow pre-screening
   • Adds support for configuring model roles via ~/.config/raptor/models.json or environment variables
   • Adds 39 Semgrep rules covering OWASPOWASPA nonprofit foundation that publishes open standards, guides, and ranked vulnerability lists—most famously the OWASP Top 10—giving cyber tools a vendor-neutral benchmark to test or report against. ASVS V11 (Cryptography) across Java, Python, JavaScript, Go, Ruby, PHP, Rust, and C
   • Adds /project subcommand for named workspaces to organize runs, merge findings, track coverage, and diff between runs

   5 details in Everything