groundwork

Agent Communication Protocol — fan-out dispatch reference for Claude plugin orchestration

Disciplined 6-phase bug diagnosis loop. Build feedback loop, reproduce, hypothesise, instrument, fix with regression test, cleanup with post-mortem. Use for all bugs — bypasses PRD entirely. Ends with advisor-gate completion gate.

Interview-based planning skill. Ask one question at a time with recommended answers, cross-reference codebase, detect and defer to project-level planning conventions, and synthesize a concise plan that feeds vertical-slice. Use for features, small changes, and bug scoping.

Run the groundwork acceptance test harness. Documents how to test the plugin locally.

MANDATORY user acceptance testing before advisor-gate. Enforces real end-user testing — TUI via tmux/expect, API via real HTTP calls in docker-compose, web apps via Playwright browser. Unit/integration tests are INSUFFICIENT unless they exercise the system exactly as end-users experience it. No exceptions.

Engage maximum parallel fan-out mode. Use when you want to work 10x faster by dispatching all independent work simultaneously to specialist agents. Triggers on "ultrawork", "ulw", "fan out hard", "go parallel".

Bootstrap skill for the groundwork workflow suite. Loaded at every conversation start. Establishes core rules, skill triggers, and the 1% escalation heuristic. ALWAYS load this first.

Mid-session codebase architecture review. All exploration runs in isolated subagents so findings never pollute main context. Surfaces shallow modules, coupling smells, and testability gaps as an HTML report. Use between tasks to retrospect on architecture without bloating the session context window.

Disciplined 6-phase bug diagnosis loop. Build feedback loop, reproduce, hypothesise, instrument, fix with regression test, cleanup with post-mortem. Use for all bugs — bypasses PRD entirely. Ends with advisor-gate completion gate.

Implementation orchestration skill. Decompose into vertical slices for maximum general-purpose fan-out, then validate behavior (not code structure). MANDATORY after a plan or interview. Use vertical-slice skill for conflict-free slice planning and the .groundwork/run.json ledger.

Interview-based planning skill. Ask one question at a time with recommended answers, cross-reference codebase, detect and defer to project-level planning conventions, and synthesize a concise plan that feeds vertical-slice. Use for features, small changes, and bug scoping.

Control another OpenCode instance via the Agent Client Protocol (ACP). Start an ACP server, send prompts (single-turn or multi-turn), capture responses. Use for testing workflows, delegating work to isolated agents, or running tasks in separate projects.

Bootstrap skill for the groundwork workflow suite. Loaded at every conversation start. Establishes core rules, skill triggers, and the 1% escalation heuristic. ALWAYS load this first.

Bootstrap skill for the groundwork workflow suite. Loaded at every conversation start. Establishes core rules and lists all available skills with triggers. ALWAYS load this first.

Decompose any implementation task into conflict-free parallel slices for maximum general-purpose fan-out. Each slice is a thin end-to-end tracer through all layers for one user-facing behavior. Use before delegating to general-purpose agents — this is the decomposition phase, not the implementation phase. Writes the .groundwork/run.json ledger that the Stop-gate hook enforces.

Executor-first workflow with advisor checkpoints at hard decisions AND mandatory gate approval before declaring any task complete. The advisor operates as a strategic technical consultant — providing deep architectural insight, trade-off analysis, and effort-aware recommendations — not just a yes/no gate. ALWAYS required before claiming done.

Build throwaway prototypes to flesh out a design before committing. Logic prototypes (interactive TUI) or UI prototypes (variant switcher). Delete when done. Use for spikes and design exploration.

Spec lifecycle management skill. Handles spec creation, revision, and deprecation. Tracks version, linked PRs/commits, and status transitions.

Executor-first workflow with advisor checkpoints at hard decisions AND mandatory gate approval before declaring any task complete. The advisor operates as a strategic technical consultant — providing deep architectural insight, trade-off analysis, and effort-aware recommendations — not just a yes/no gate. ALWAYS required before claiming done.

Stateful time-bounded improvement loop with evaluator contract and dual logging

When context is getting long or a fresh session is needed, present the user with a choice between same-session continuation (summary + continue) or /handoff to create a new session. Never silently drop context.

Build throwaway prototypes to flesh out a design before committing. Logic prototypes (interactive TUI) or UI prototypes (variant switcher). Delete when done. Use for spikes and design exploration.

Hand off the current session to a freshly spawned Claude Code session. Writes a structured handoff document referencing this session's transcript, then spawns (or prints the launch command for) a successor session. Use when the context window is filling up, when switching machines or terminals, or when ending a work block with a clean baton pass.

Manage persistent project goals that survive context compression and session restarts. Set objectives with acceptance criteria, check status, pause, resume, or clear. The active goal is injected into every message as a reminder.