Fabricated references
Checks whether pinned GitHub Action commits and exact package versions resolve to real upstream objects instead of trusting a plausible-looking string.
Open-Source Security Engineering
Supply chain defense for the AI-assisted stack.
depfence statically inspects dependencies, workflows, model files, and agent configuration for threats conventional CVE scanners were not designed to see.
$ depfence scan .
discovering lockfiles, workflows, model files, and agent configuration
LOCAL source inspection stays on this machine
STATIC package code is not executed
VERIFY dependency versions and GitHub Action references
INSPECT prompt injection, MCP, agent skills, and CI trust boundaries
REPORT table · JSON · SARIF · CycloneDX · SPDX
Illustrative command flow. Findings depend on the repository being scanned.
Security gaps worth engineering for
depfence complements established vulnerability scanners by covering the control plane around AI-assisted software development.
Prompt injection
Inspects source, metadata, commits, and configuration for instructions intended to manipulate AI-assisted review and development workflows.
MCP and agent configuration
Looks for tool shadowing, credential exposure, suspicious instruction fetches, mutable definitions, and unsafe agent configuration.
CI/CD trust boundaries
Audits GitHub Actions for expression injection, excessive permissions, untrusted artifacts, mutable actions, and privileged workflow paths.
Model and build formats
Examines model files, build hooks, manifests, and project metadata for executable formats, unsafe deserialization, and install-time behavior.
Evidence-ready outputs
Produces human and machine-readable results for local review, CI gates, GitHub code scanning, and software bill-of-materials workflows.
Inspect without executing.
Dependency security tooling should not become another execution path. depfence analyzes source and metadata statically; package code is not run, and source code is not sent to an analysis service.
Local source inspection
No package execution
Offline mode available
Explicit network behavior
Designed to fit existing workflows
One scanner surface, several ways to consume the evidence.
Analysis surfaces
Dependencies
GitHub Actions
MCP
Agent Skills
Model Files
Build Hooks
Outputs and integration
CLI
GitHub Actions
MCP Server
JSON
SARIF
CycloneDX
SPDX
Engineering proof, not a pitch
Security claims should carry their limits.
depfence is not a replacement for every security scanner. Established tools have broader vulnerability databases and deeper container-image coverage. Its role is to cover AI, agent, dependency, and workflow attack classes that sit outside traditional CVE matching.
The project is currently installed from source rather than PyPI. The repository documents network behavior, offline operation, limitations, and the tests behind its claims.
See the implementation, not a brochure.
depfence is public because security engineering is easier to trust when the source, limitations, and verification work are visible.