Add OpenCode for Agentic Programming Using the Plugins Superpowers, Ponytail and MCP Context7 for Harnesses

The repository increasingly relies on AI agents to write code, but without a standard agent harness every contributor picks their own tool and workflow: different assistants, no shared process, and no common source of up-to-date library documentation. The result is inconsistent output quality, workflows that cannot be enforced, and agents reasoning from stale API knowledge. Which harness and plugins should we standardize on for agentic programming?

Chosen option: "Adopt OpenCode with the superpowers, ponytail and MCP Context7" (because it is a CLI-first harness that standardizes agentic programming on top of an enforceable, plugin-driven workflow). Positive consequences: superpowers enforces an agent workflow for creating and debugging code; ponytail steers agents toward minimal code, guarding against over-engineering; and MCP Context7 keeps library and framework documentation current; all three are registered through OpenCode’s plugin and MCP configuration. Negative consequences: the workflow guidance lives in third-party repositories that we must track, plugin behavior can shift between OpenCode releases, and the harness adds a workflow overhead that some tasks do not need.

OpenCode is the best option for agentic programming, because it is local, configurable via plugins and MCP servers, and its skills integrate directly with the repository’s own commands and ADRs. If another assistant was chosen, the workflow could not be versioned with the repository. If we kept ad hoc prompts, nothing would prevent each agent session from drifting back to inconsistent behavior.