reflect

Run /reflect with no arguments for a broad review, or /reflect followed by a focus area like a release workflow. The skill inspects recent sessions, project notes, and existing assets, then reports repeated patterns with evidence and frequency.

Choose the smallest useful form: a prompt rule for small behavior changes, a skill for reusable task guidance, a command for stable manual triggers, and a custom agent only for a distinct specialist lane. Reflect scores candidates and recommends the least powerful option that solves the problem.

No. Reflect presents a concise proposal and asks for confirmation before writing files or changing config, prompts, agents, skills, or MCP permissions. When changes are applied, it reminds you to restart OpenCode for them to take effect.

It uses the current conversation, project-local guidance like AGENTS.md and .opencode notes, existing skills and agents, oh-my-opencode-slim configuration, and recent OpenCode logs when safe to inspect. It avoids private files and credentials unless explicitly requested.

Avoid it for ordinary implementation work, one-off debugging, broad architecture reviews, or speculative agent creation without workflow evidence. Candidates seen only once or already covered by existing assets are skipped.