ship-mate — agentic threat model

Uses Claude Code (Anthropic Claude models) as its foundation. Risks include prompt injection via malicious user stories or codebase files, leading to unauthorized code generation or execution.

Not certain from the listing — operates on local codebase files and user-provided story files. Lacks explicit mention of vector databases or RAG, but codebase context acts as the primary data ingestion layer.

Orchestrates a multi-agent pipeline (orchestrator, architect, developer, reviewer, QA). Vulnerable to tool misuse and insecure tool integration, as the framework translates model outputs directly into file writes and PR creation.

Runs locally or in CI/CD environments to execute Playwright tests and edit code. This creates severe risks of container/host compromise or lateral movement if malicious code is injected and executed during the test phase.

Not certain from the listing — there is no mention of built-in guardrails, evaluation frameworks, or real-time observability to detect anomalous file modifications or malicious test scripts before execution.

Not certain from the listing — security controls depend entirely on the host environment's git/SSH permissions and Claude Code's underlying configuration. No native authentication or policy enforcement is described.

Features a highly integrated multi-agent pipeline (orchestrator -> architect -> developer -> reviewer -> QA). Vulnerable to cascading failures and trust abuse where a compromised developer agent bypasses the reviewer and QA agents.