Favot Corentin

You can tune OpenCode agents, models, prompts, MCPs, and presets; generate hierarchical codemaps of unfamiliar repositories; clone dependency source for internals inspection; manage Git worktrees as isolated coding lanes; plan verification for non-trivial changes; orchestrate large multi-phase efforts; and simplify code without changing behavior.

Software engineers and web developers using OpenCode who face recurring configuration friction, need to onboard onto unfamiliar codebases, debug library internals from source, or run risky parallel changes in isolated Git worktrees with a credible verification path before merging.

Start with codemap to map an unfamiliar repo, use clonedeps to read dependency source, open worktrees for isolated lanes, run verification-planning before implementing features or refactors, apply deepwork for large multi-phase efforts, then finish with simplify for readability and reflect to capture reusable patterns.

Avoid codemap unless explicitly asked for repository mapping since it is expensive; skip clonedeps for ordinary documentation questions; do not activate deepwork for routine multi-file changes; and only run simplify after behavior is fully understood, since it must not alter functionality.

You need an OpenCode environment where skills, agents, MCPs, and presets can be configured, plus a Git repository for worktree isolation. clonedeps requires network access to clone dependency repositories into an ignored local workspace for source-level inspection.