Opencode Coding is a skill published by agensi that imposes a structured engineering workflow on AI coding agents. Rather than generating speculative code changes, it mandates a three-part contract: reproduce and verify failures before touching code, implement the narrowest defensible change, and report exactly what was verified, what was inferred, and what remains unknown.
The skill is framework-agnostic, meaning it can be applied to any tech stack without configuration changes. Its evidence-based debugging approach requires the agent to identify root causes and reproduce failures before proposing a fix, reducing the risk that a patch addresses symptoms rather than the underlying problem.
Minimal-impact diffs are a core constraint: the skill directs the agent to preserve existing project patterns and limit the scope of every change, which lowers regression risk in established codebases. A verification-first workflow requires running targeted tests, linters, or manual validation steps before a task is declared complete, rather than relying on visual inspection of generated code.
Every completed task produces a standardized Response Contract output that explicitly labels which conclusions are verified, which are inferred, and which are still unknown. This makes the agent’s confidence level transparent and auditable.
The skill is intended for complex feature implementation, surgical bug fixing, refactoring legacy modules, and PR reviews where functional correctness must be demonstrated rather than assumed.