The Problem

Modern agentic codebases generate CI/CD misconfigurations, hard‑coded secrets, and injection vectors at speed. A single overlooked permission or leaked key can become an entry point for supply‑chain or LLM‑based attacks. Teams need a fast, parallel scan that surfaces not only code flaws but also the operational risk of running autonomous agents in production.

What This Does

ship-safe is a CLI security scanner that runs 22 specialized agents in parallel against a codebase. Key capabilities, mapped to actual source files, include: Secret detection – cli/utils/secrets-verifier.js probes provider APIs to verify whether leaked keys are still active. MCP tool injection – cli/agents/mcp-security-agent.js and cli/hooks/pre-tool-use.js/ post-tool-use.js inspect agent‑tool interactions for command‑injection patterns. CI/CD pipeline poisoning – cli/agents/cicd-scanner.js and .github/workflows/ci.yml audit GitHub Actions for unsafe steps, container privileges, and secret exposure. LLM/agentic AI security – cli/agents/llm-redteam.js, ai-defense/prompt-injection-patterns.js, and ai-defense/system-prompt-armor.md map to the OWASP Agentic AI Top 10 (ASI‑01 – ASI‑10). Supply‑chain and SBOM – cli/agents/supply-chain-agent.js and cli/utils/plugin-loader.js generate a software bill of materials and flag vulnerable dependencies from the OSV feed. Red‑team scoring – cli/commands/red-team.js triggers the full 80‑attack‑class suite; results feed the Findings Dashboard (web UI) and can be exported as GitHub issues.

The platform also provides an Agent Studio (UI) for building custom Hermes agents, an Agent Console (SSE chat with ANSI rendering), and Agent Teams that orchestrate a lead agent, parallel delegates, and an executive report.

How To Use It

StepCommand (from README)Notes
Installnpm install -g ship-safe (or npx ship-safe)Pulls the package from npm; package.json at repo root lists the entry point cli/index.js.
Basic auditnpx ship-safe audit .Scans the current directory with all 22 agents, pulls the live OSV advisory feed, and outputs a JSON findings file.
Deep analysisnpx ship-safe audit . --deepEnables LLM‑powered verification (Anthropic, OpenAI, Google, Ollama). Requires API keys stored in .env or provided via environment (ANTHROPICAPIKEY, OPENAIAPIKEY).
Agentic loopnpx ship-safe audit . --agentic <br> npx ship-safe audit . --agentic 5 --agentic-target 85Scans, auto‑annotates fixes, re‑scans until the security score ≥ 75 (configurable).
Red‑teamnpx ship-safe red-team .Runs the full 80+ attack class suite for a comprehensive breach‑simulation.
Changed‑file scannpx ship-safe diff <br> npx ship-safe diff --stagedFast pre‑commit/PR scan limited to modified files (uses git diff).
CLI entry pointcli/bin/ship-safe.jsDirect execution without npx after global install.

Configuration – No mandatory env vars for a pure code‑scan. For deep/agentic runs set the relevant LLM keys; the project ships webapp/.env.example showing ANTHROPICAPIKEY, OPENAIAPIKEY, GOOGLEAPIKEY. The cli/providers/llm-provider.js reads these values and falls back to a local mock if absent.

Real‑World Use

A security engineer integrates ship-safe into the CI pipeline:

.github/workflows/ci.yml (excerpt) steps: uses: actions/checkout@v4 run: npx ship-safe audit . --deep name: Upload findings uses: actions/upload-artifact@v3 with: name: ship-safe-findings path: findings.json

The workflow fails on any critical or high‑severity finding, prompting developers to remediate before merge. After fixing, the next run re‑evaluates the score, providing a measurable security‑posture trend.

Code Health & Issues

Tests: 3 test files exist (cli/tests/agents.test.js); coverage is limited. Adding unit tests for each agent’s pattern‑matching logic would improve confidence. CI: GitHub Actions workflow (.github/workflows/ci.yml) runs on push and pull‑request; it currently lints and type‑checks but does not invoke ship-safe scans. Extending the workflow to run npx ship-safe diff --staged would catch issues earlier. License & Security: LICENSE (MIT) and SECURITY.md are present, meeting open‑source hygiene. No hard‑coded secrets observed in the repo root, but webapp/.env.example should be kept out of version control. Dependency hygiene: package-lock.json is vendored; periodic npm audit is recommended to catch transitive vulnerabilities. Documentation: README.md, CHANGELOG.md, and per‑agent SKILL.md files under claude-code-plugin/skills/ provide clear usage guides. The web dashboard docs are hosted externally (shipsafecli.com/docs).

No structural red flags (missing package.json, broken CI, absent license) were detected.

The Bottom Line

ship-safe delivers a broad, parallel security audit for agentic codebases with minimal setup—just npm install -g ship-safe and an optional .env for LLM deep‑analysis. Its strength lies in the extensive agent suite, live OSV feed, and the new Agent Studio/Teams UI that turns raw findings into actionable security ops. The main trade‑off is that full deep‑analysis requires API keys, and test coverage is sparse; teams comfortable adding CI integration and modest test expansion will get the most value. It is well‑suited for midsize‑to‑large development organizations that run autonomous agents or heavily consume third‑party AI tools, while solo developers may find the feature set heavier than needed.