Open-source application security projects, mostly with an AI focus.
appsec-analyzer is a command-line tool for threat assessments of local and remote repositories. A deterministic controller runs the assessment and integrates the LLM: Claude by default, Codex and other providers experimentally.
- Combines deterministic scanners, a STRIDE-based threat survey, a catalog-based verifier, and abuse case checks. You can add your own abuse cases.
- Writes a Markdown report and a YAML data file, and exports to PDF, HTML, and other formats.
- Opens the findings in Claude Code or Codex for further analysis.
Excerpt from an OWASP Juice Shop assessment, created with appsec-advisor: the architecture derived from the repository, with findings per component and the assets each layer stores.
appsec-analyzer-examples has example reports for OWASP Juice Shop and DVWA as Markdown and YAML files, with all diagrams as SVG.
appsec-verifier is a command-line tool that tests whether the findings of an appsec-analyzer assessment hold up in a running application you are authorized to test. It uses Claude by default, Codex and GitHub Copilot as alternatives.
- An AI agent plans the tests, but only a check in code confirms a finding, never the model's verdict alone.
- Tests cross-site scripting, access control, injection, path traversal, open redirect, and SSRF. Every request stays within the scope you define.
- Shows the requests behind each verdict, with evidence images and a HAR file.
Evidence image from a run against OWASP Juice Shop. A single quote in the search parameter q causes a database error and two quotes do not, so the Verifier confirms the SQL injection.
The Juice Shop example has the complete run with console log, report, and all evidence images.
appsec-tools has standalone tools that run deterministically, without an LLM.
-
Security Score gives a repository a score from 0 to 100 and a grade from A to F, based on static checks. For a change, it rates the security relevance instead.
Excerpt of
security-score --detailson OWASP Juice Shop, which is insecure on purpose. "partial" means that some topics could not be assessed, so there is no grade. -
Requirement Harvester reads requirements from websites and wikis and writes them to an OSCAL catalog in YAML.
aiscb is a set of secure-coding rules for Claude Code, GitHub Copilot, OpenAI Codex, and Kiro.
- A compact core stays in the assistant's context on every prompt.
- Topic modules such as web, authentication, secrets, or supply chain load only when a task needs them.
- The assistant applies the rules while it writes code and reports remaining risks in a security note.
The same prompt in Claude Code, without and with aiscb. Without the baseline, Claude writes a custom cipher. With it, Claude cites the rule against hand-rolled crypto and asks what the code is for.
appsec-advisor is a Claude Code plugin for application security. Most of it has moved into appsec-analyzer, and the plugin is being rebuilt.
-
appsec-advisor-examples has example threat models for OWASP Juice Shop and other deliberately vulnerable apps.
-
appsec-advisor-packaging-template is a template for packaging appsec-advisor as an internal plugin with your organization's configuration.
Example rollout at Acme: an internal repository packages a pinned appsec-advisor release with the organization profile. Developers install the result and create threat models with Claude Code.
appsec-skills is a collection of application security skills for coding assistants, including:
- create-threat-model creates, updates, or reviews a threat model from requirements, design documents, and team knowledge, even before code exists.
- analyze-threats examines features, changes, pull requests, or design decisions for threats.
tss-web is an open security requirements framework for web applications and services.
- Covers technical and organizational controls for developing and operating web applications.
- Sits between high-level security policies such as ISO/IEC 27001 and technology-specific secure-coding guidelines.
- No longer maintained.
Questions and bug reports are welcome in each project's issue tracker.
Maintainer: Matthias Rohr
LinkedIn