yocto-security-tools

CVE management tools for Yocto/OpenEmbedded Linux distributions. They find the upstream commits that fix a CVE, apply them to your recipes, and optionally use an AI backend to resolve the conflicts and build failures that follow.

How it works

graph LR
    E["cve-metadata-extractor<br/>Find fix commits"] -->|cve-metadata.json| C["cve-corrector<br/>Apply patches via devtool"]
    C -->|exit code + state| A["cve-agent<br/>AI-assisted resolution"]
    A -->|subprocess| C

Loading

Each tool works standalone. Chain them with --cve-info cve-metadata.json.

Requirements

• Python 3.10+ and Git
• A sourced Yocto build environment (BBPATH set) for cve-corrector and cve-agent
• An AI backend for cve-agent — see Modules below

Installation

pip install yocto-security-tools

From source:

git clone https://github.com/Ericsson/yocto-security-tools.git
cd yocto-security-tools
pip install -e .

Quick start

# 1. Find fix commits for the CVEs in a Yocto CVE summary
cve-metadata-extractor --yocto-summary cve-summary.json --output cve-metadata.json

# 2. Source your Yocto build environment
source oe-init-build-env

# 3. Apply one fix
cve-corrector --cve-id CVE-2024-1234 --cve-info cve-metadata.json

# ...or let an AI backend resolve conflicts and build failures for you
cve-agent --cve-id CVE-2024-1234 --cve-info cve-metadata.json

Modules

cve-metadata-extractor

Finds the commits that fix a CVE by querying Debian security-tracker, OSV, CVEList V5, the Ubuntu CVE Tracker, and NVD, then writes a single cve-metadata.json for the other two tools. Accepts a Yocto cve-summary.json (--yocto-summary) or explicit CVE IDs (--cve-id). Optionally checks whether a fix already landed in an OpenEmbedded branch (--check-oe).

→ Full reference

cve-corrector

Applies a fix to a recipe using devtool: cherry-picks the upstream commit into the recipe's source tree, builds, runs ptest, and finishes the change into a layer. Stops with a specific exit code when it needs help — conflict, build failure, or ptest failure — so you can fix it by hand and resume with --continue. --fix-url is repeatable and applies two or more commits as one ordered, dependent chain.

→ Full reference

cve-agent

Runs cve-corrector as a subprocess and, on a recoverable exit code, starts a guarded AI session to resolve the conflict or failure, then retries. Backends are interchangeable via --backend:

Backend	--backend	Needs
Kiro CLI	kiro (default)	kiro-cli
Claude Code	claude	Authenticated claude CLI on PATH
Native OpenAI-compatible	openai / openai-<profile>	A tool-capable OpenAI-compatible endpoint, including local Ollama
Custom plugin	your own name	A file in extra/ implementing AIBackend

Every backend runs under the same file-scope guard, so the AI can only modify the files the upstream fix touches. Check a backend is installed and responding with cve-agent --backend <name> --verify-backend. Use --cve-list for batch runs.

→ Full reference · OpenAI-compatible/Ollama setup

Documentation

docs/README.md indexes everything: per-tool references, configuration, and the design docs covering the result schema, agent artifacts, preflight checks, the corrector-to-agent handoff, safe patch transfer, semantic security validation, and the evaluation harness.

Plugins

Add a CVE data source or an AI backend by dropping a .py file into extra/ — no existing file needs to change. See extra/README.md for the plugin guide.

Configuration

Data and cache directories follow the XDG base directory spec and are overridable, as are the extractor's config path and the API tokens. See docs/configuration.md.

Development

python3 -m venv venv
source venv/bin/activate
pip install -e ".[dev]"
pytest

See CONTRIBUTING.md for full development guidelines.

License

MIT — see LICENSE