Independent security researchers have issued an urgent advisory urging developers to discontinue use of OpenCode, a widely adopted open-source AI coding agent boasting over 161,000 GitHub stars. Following a comprehensive architectural and security audit, the analysis identifies systemic vulnerabilities and functional deficiencies that fundamentally undermine the tool safety and reliability in production environments. Functionally, OpenCode demonstrates severe deficiencies in context management and agent orchestration. The platform mishandles prompt caching, leading to excessive computational overhead and latency during long sessions. Its built-in pruning and compaction mechanisms operate inefficiently, often degrading performance rather than optimizing it. Furthermore, subagent interactions suffer from broken state persistence and flawed permission routing, causing frequent execution halts and data loss. Documentation and tool interfaces remain incomplete and poorly engineered, compounding usability challenges. The most critical findings center on OpenCode default security posture. The agent is configured to establish remote model connections immediately upon launch, transmitting local directory contents and file metadata to external servers without explicit user configuration. Default web-fetching capabilities and unrestricted internet access further expose systems to prompt-injection and remote code execution risks. Security controls, particularly command-line permission filtering, rely on brittle text-based parsing rather than sandboxed execution. Researchers demonstrated that these filters can be trivially bypassed using shell redirections, environment variables, and non-standard command syntax. Additionally, an on-disk permission persistence feature grants permanent execution rights to approved commands, creating long-term escalation pathways. Path validation logic exhibits similar flaws. While OpenCode attempts to restrict file access to the current working directory or active Git repository, its AST-based path resolver frequently fails to trigger prompts for cross-directory reads or writes. Commands absent from its predefined file-access list bypass validation entirely. The platform self-upgrade mechanism relies on downloading and executing raw scripts via unverified curl commands, perpetuating a dangerous supply-chain risk. This pattern aligns with historical vulnerabilities, including a previously disclosed default HTTP server configuration that exposed local machine shells to remote exploitation, and an authentication command capable of fetching and executing arbitrary user-supplied URLs. Proponents often suggest containerization via Docker as a mitigation strategy, but security auditors reject this as an inadequate workaround that shifts responsibility away from native security-by-design principles. The advisory emphasizes that coding agents require robust, OS-level isolation and strict capability boundaries rather than heuristic command filtering. The analysis also contextualizes broader concerns regarding local language model integration, noting that current AI-driven code generation practices frequently destabilize architectural integrity, introduce hallucination-driven bugs, and exacerbate technical debt. In light of these findings, the research community strongly recommends immediate suspension of OpenCode deployments until core architectural flaws, permission models, and default network behaviors are rigorously remediated. Developers are advised to adopt verified, sandboxed alternatives and implement strict network egress controls until the platform addresses these critical security gaps.