Security skills & reviewers
Use Beam’s packaged review workflows for skills, MCP servers, AI assets, application boundaries, and incident evidence.
On this page
The Beam CLI package includes a Skills collection of focused security workflows and seven paired specialist reviewer profiles. These are instructions for your assistant, not an automatically enabled runtime security layer. Review and install the subset relevant to your project.
Choose a workflow
The collection covers skill and MCP review, AI assets and model artifacts, prompt injection, agent permissions, secrets, dependencies, web/API boundaries, containers, cloud, CI/CD, monitoring, and incident response. security-assessment is the parent workflow for a broader assessment.
Use a concrete scope: identify the exact artifact, its source, the question you need answered, and what actions are permitted. A static review should report untested runtime behavior explicitly.
Locate the packaged files
The npm package is @agent-beam/beam and ships Skills alongside its built CLI. Use npm root -g to locate globally installed packages, then inspect the collection's README and manifest. You can also work from a reviewed Beam CLI source checkout.
Install selected skill folders into the skill directory configured by your client. Files left under the package's Skills directory are not automatically discovered by every assistant. Refresh client discovery after installation.
Install a project reviewer
From a reviewed CLI checkout, this example adds one Claude Code profile to that project without overwriting an existing file:
mkdir -p .claude/agents
cp -n Skills/agents/claude/beam-skill-auditor.md .claude/agents/
For Codex:
mkdir -p .codex/agents
cp -n Skills/agents/codex/beam-skill-auditor.toml .codex/agents/
Use the actual destination project path when installing elsewhere. Review intentional updates rather than assuming a no-clobber copy replaced an older profile.
Pick a specialist
| Reviewer | Focus |
|---|---|
beam-skill-auditor | Skill artifacts, referenced code, and installation permissions |
beam-mcp-auditor | Server identity, configuration, tools, and connection scope |
beam-ai-auditor | AI assets, instructions, retrieval, memory, and models |
beam-appsec-auditor | Web, API, authentication, secrets, dependencies, and CI |
beam-infrastructure-auditor | Containers, cloud roles, deployment, and isolation |
beam-incident-analyst | Supplied activity and evidence of proposed versus completed effects |
beam-findings-reviewer | Evidence quality, severity, duplicates, and coverage |
Profiles inherit host capabilities and model choices. A read-only tool list or sandbox setting alone does not define an OS/network boundary. Provide parent-generated scan results and hashes when the reviewer cannot run those tools, and keep live probes or changes outside a read-only assessment.