leoncheng57

Draw annotated ASCII diagrams instead of prose walls when explaining a multi-step plan, a large change set, a data flow, a UI proposal, or…

Run a sustained multi-wave build in one OpenCode session, using background subagents to compress research while the parent keeps…

Escalate one large research question to parallel read-only subagents instead of grinding through it inline, then synthesise the reports…

Choose the right visual medium when writing documentation and use the diagram and preview tooling actually installed on this machine…

Wrap every assistant reply in lines of duck emojis.

Hand the request the user just made to a background subagent with the task tool and return immediately instead of doing the work inline.

Launch one standalone OpenCode session from another and explicitly carry over its agent or mode, model, reasoning effort, working…

Generate a short, executable checklist for a human to verify completed user-visible behaviour across browser UI, CLI/scripts…

Interview the user in rounds about a plan, design, or decision, working a design tree until every branch is settled and nothing is left…

Coordinate implementation waves through separate OpenCode TUI workers in isolated git worktrees and cmux workspaces.

Research several independent feature ideas in parallel with read-only subagents, compile the findings into self-contained handoff prompts…

Argue against a plan, design, or diff the agent just produced, switching sides and grounding every objection in file:line, command output…