# anthropics/defending-code-reference-harness

Skills for threat modeling, scanning, triage, patching, plus an autonomous scanning harness you can /customize

Repository: https://github.com/anthropics/defending-code-reference-harness
Canonical: https://ross.abutalabs.com/products/defending-code-reference-harness
Homepage: https://claude.com/blog/using-llms-to-secure-source-code
Language: Python
License: NOASSERTION
License Family: other
Topics: security
Last push: 2026-08-06T21:42:09+00:00

## Health v2 (maintenance only)
Score: 57/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 96, release rhythm 35, longevity 7
- inputs: {"age_days": 103, "days_push": 27, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, young, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 7368, forks 594 (observed 2026-08-28T04:09:58.517316+00:00)

## What it is
A reference implementation from Anthropic for autonomous vulnerability discovery and remediation using Claude, including Claude Code skills for threat modeling, scanning, triage, and patching, plus an autonomous scanning harness configured for C/C++ memory vulnerabilities. It also includes detection-and-response skills for hunting attackers already present in logs.

## Use cases
- find memory safety vulnerabilities in C/C++ codebases with an LLM agent
- build a threat model for my repository before scanning
- triage and verify LLM-discovered vulnerability findings to reduce false positives
- automatically generate patches for confirmed vulnerabilities
- customize an autonomous vulnerability scanning pipeline for my language or vuln class
- hunt for signs of an attacker already in my logs and propose a response

## When to choose
- you want an open-source, customizable reference pipeline for LLM-driven vulnerability discovery and remediation
- you use Claude APIs (including Bedrock, Vertex, or Azure) and want to build your own security scanning workflow
- you need skills for interactive threat modeling, scanning, triage, and patching in Claude Code
- you want to learn Anthropic's best practices for securing source code with LLMs

## When to avoid
- you need a maintained, production-ready product with support - the repo explicitly is not maintained and accepts no contributions
- you want a turnkey scanner that works on every codebase out of the box - the harness is a reference, not a product
- you prefer a managed hosted solution - consider Anthropic's Claude Security product instead
- you need a permissive license - the license is non-standard (NOASSERTION)

## Facets
- artifact type: framework
- maturity: maintenance
- function: security, vulnerability-scanning, agent-framework, penetration-testing, developer-tools
- domain: security, developer-tools, artificial-intelligence
- platform: python, cli, cross-platform
- tags: vulnerability-discovery, threat-modeling, triage, patching, claude-code-skills, llm-security, memory-safety, asan, detection-and-response, reference-implementation, ai-agents, docker

## Member repositories
- anthropics/defending-code-reference-harness (main) score 57

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:58.517316+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:38:23.873226+00:00, confidence not recorded.
  - readme: https://github.com/anthropics/defending-code-reference-harness (fetched 2026-08-28T04:09:58.517316+00:00, sha 989bec13ca56)
  - homepage: https://claude.com/blog/using-llms-to-secure-source-code (fetched 2026-08-29T08:33:50.566403+00:00, sha aa0c99625569)
  - site_page: https://platform.claude.com/docs (fetched 2026-08-29T08:33:50.578261+00:00, sha 42493d4b28e3)
  - site_page: http://claude.com/pricing (fetched 2026-08-29T08:33:50.576364+00:00, sha 71972a93716a)
  - site_page: https://claude.com/pricing (fetched 2026-08-29T08:33:50.580148+00:00, sha 54ae70c3efc4)
- Data as of 2026-08-30T08:39:29.467469+00:00.