Security Review for Claude Code Projects
Claude Code can plan, write, and test large features from your terminal. It also ships tooling such as the `/security-review` command and security plugins. Those catch common patterns. We review what they cannot: your business logic, your permission model, and the agent configuration in your repo.
30-minute intro call · fixed-scope quote · human review, not scanner output. Read client case studies.
What we find most often in Claude Code projects
Risks fall into two groups: issues in the application Claude Code helped build, and issues in the agent setup that influences how it builds.
Business logic automated review misses
Automated security review is good at known patterns such as injection or hardcoded secrets. It cannot tell whether a trial user should be able to export data or whether a refund flow can be abused.
CLAUDE.md and hooks as an attack surface
CLAUDE.md files guide the agent and hooks run shell commands automatically. In a cloned or shared repo, either can be used to inject instructions or execute code on a developer's machine.
Over-broad permission settings
Permission modes and allowlists that let the agent run any command or edit any file trade safety for speed. Combined with untrusted input, that can lead to unintended actions.
Unvetted MCP servers, plugins, and skills
Third-party MCP servers, plugins, and skills can read context, call tools, and access credentials. Each one is a dependency that should be reviewed like a package.
Authorization drift across long sessions
Multi-step agent sessions can implement a feature end to end while skipping an ownership check in one route. Tests written in the same session often miss the same case.
Secrets read into context
If the agent can read `.env` files or cloud credentials, those values can end up in logs, commits, or generated code. Access should be limited to what the task needs.
What a Claude Code security review covers
A fixed-scope review of your application and the Claude Code configuration committed to your repository.
Checks you can run yourself
Start here before you book anything. If any of these fail or you are not sure how to check, that is the signal to get a second pair of eyes.
- Run `/security-review` on your current changes and treat the output as a starting point, not a sign-off.
- Read every CLAUDE.md and hook in the repo, especially in projects you cloned, and remove anything you did not write.
- Review your permission settings and require confirmation for shell commands in unfamiliar repositories.
- List the MCP servers and plugins you have enabled and remove any you no longer use.
- Pick your most sensitive endpoint and confirm it checks ownership, not just login.
Want a scored version? Take the free vibe code security check or work through the 25-point launch checklist.