OpenAgents Control (OAC)
AI agents that learn your coding patterns and generate matching code every time, with approval gates for controlled execution.
Evidence shows approval gates emphasized, but no actual least-privilege implementation details. User confirmation mechanism described in README but lacks code-level verification. Data flow transparency insufficient; how context files are processed not explained. Sensitive data handling not mentioned. Dependency security lacks dependency list or audit. External effects (e.g., network requests) not explicit. Rollback mechanism not mentioned. Source attribution only via author info, but publisher unverified.
Self-consistency good; README and package.json descriptions align. Dependency availability lacks full dependency list; only devDependencies has glob. Failure messages lack concrete error handling examples.
Audience clear (developers, teams), scenarios diverse. Capability boundaries explained in README (e.g., no parallel execution). Trigger precision insufficient; command trigger conditions not detailed. Environment fit good; supports multiple languages and models.
Information architecture clear; README, SECURITY.md present. Install notes detailed. Naming stability not explicit for version compatibility. Examples and FAQ rich. Known limitations mentioned in README. License MIT. Versioning and changelog supported by workflows. Maintenance responsibility not explicit, but author info exists.
Output usability high; example workflows provided. Marginal value clear; solves pattern matching problem. Cost-benefit has claimed token reduction but lacks independent verification.
Claim traceability insufficient; no test results or evidence provided. Cross-source corroboration limited; relies on README only. Fact-inference separation unclear; many performance claims unverified.
- Publisher identity unverified; treat with caution.
- Performance claims (e.g., 80% token reduction) lack independent verification.
- Dependency security lacks full manifest; audit needed.
- Data flow and sensitive data handling opaque; further review required.
What does this agent do, and when should you use it?
OpenAgents Control (OAC) is an AI agent framework built on OpenCode, designed for plan-first development workflows. It uses a context system to load project-specific coding patterns before code generation, and approval gates to ensure every modification is human-approved. The framework includes multiple editable agents (OpenAgent, OpenCoder, SystemBuilder) and specialized subagents (ContextScout, TaskManager, TestEngineer, CodeReviewer, BuildAgent, DocWriter, ExternalScout). Its core MVI (Minimal Viable Information) principle reduces token usage significantly, supports multi-language and model-agnostic operation, avoiding vendor lock-in.
OAC operates through the OpenCode CLI. Users invoke agents via commands like opencode --agent OpenAgent, where the agent uses ContextScout to discover relevant context files, proposes a plan based on those patterns, and waits for approval before executing code changes, tests, and reviews. Specialized subagents handle specific tasks: TaskManager decomposes features, CoderAgent implements code, TestEngineer writes tests, CodeReviewer performs security reviews, BuildAgent runs type checks and builds, DocWriter generates documentation, and ExternalScout fetches live docs for external libraries. Agent behavior is configured via editable Markdown files, and commands like /add-context, /commit, /test, /optimize enhance workflows.
- Development teams with established coding standards who want consistency across all generated code.
- Developers needing production-ready code that ships without heavy refactoring.
- Users who want human oversight in AI-driven development through approval gates.
- Cost-conscious users aiming to reduce token consumption with MVI principles.
- Teams using multiple AI models (Claude, GPT, Gemini) to avoid vendor lock-in.
What are this agent's strengths and limitations?
- Context system ensures code matches your patterns, reducing refactoring effort.
- Approval gates provide human control over AI actions, preventing surprises.
- MVI principle cuts token usage dramatically (example: 8000 to 750 tokens), saving costs.
- Editable agents in Markdown allow customization without vendor lock-in, supporting any model.
- Requires OpenCode CLI installation, which may be an extra step for some users.
- Approval workflow slows down execution compared to autonomous agents.
- Initial context setup takes time (10-15 minutes) and requires ongoing updates.
- Multi-language support is less battle-tested for Python, Go, Rust compared to TypeScript/Node.js.
How do you install or deploy this agent?
Prerequisites: OpenCode CLI (free, open-source), Bash 3.2+, Git. One-command install: curl -fsSL https://raw.githubusercontent.com/darrenhinde/OpenAgentsControl/main/install.sh | bash -s developer. Interactive install: curl -fsSL .../install.sh -o install.sh && bash install.sh. Update: curl -fsSL .../update.sh | bash.
How do you use this agent?
After installation, start an agent with opencode --agent OpenAgent and enter requests like "Create a user authentication system". The agent analyzes, proposes a plan, and executes step-by-step after approval. For best results, run /add-context to add your coding patterns (10-15 minutes).
How does this agent compare with similar options?
Compared to Cursor/Copilot, OAC offers editable agents, approval gates, context system, and MVI token efficiency; compared to Aider, it adds team patterns, approval workflow, and smart pattern discovery; compared to Oh My OpenCode, it focuses on control and repeatability over autonomous parallel execution.
FAQ
Do I need to install per project?
.opencode/) is recommended for team sharing via git; global install (~/.config/opencode/) works for personal defaults.How do I add or update team patterns?
/add-context interactive wizard answering 6 questions. Update with /add-context --update as your project evolves.