← ClaudeAtlas

boi-delegationlisted

Delegate tasks to BOI v2. Covers TOML spec authoring, dispatch, context wiring, and monitoring. Use when delegating any non-trivial task — code changes, research, generation, or multi-step work.
mrap/hex · ★ 2 · AI & Automation · score 81
Install: claude install-skill mrap/hex
> **Updated 2026-05-24 for BOI v2 cutover.** > v1's SKILL.md is preserved in git history if you need to see what changed. # BOI Delegation ## BOI Is the Default — Always BOI is the default for ALL non-trivial work. Not just code — research, analysis, creative synthesis, brainstorming, everything. BOI v2 provides what inline work cannot: - **Plan + critique phases** — every spec passes through `plan` and a cross-model `critique_plan` before any code runs. - **Deterministic verify gates** — `workspace_verify_in` / `workspace_verify_out` bracket every task; verifications run in a clean subshell. - **DAG-aware scheduling** — `blocked_by` is honored, dependent tasks fan out in parallel as predecessors pass. - **Worktree-per-task isolation** — each task gets its own git worktree; no cross-task contamination. - **Typed reasons everywhere** — every state transition carries a typed reason (cancel, fail, block), enforced at the recovery commands. **The only exception:** subagents (Agent tool) for quick bounded questions where you need the answer *right now* in this conversation and the full answer fits in context. Examples: "summarize this file", "what's the status of X", "parse this output." If in doubt, BOI. The plan/critique stage and verify gates justify the async cost. ## Pre-Second-Edit Gate (MANDATORY) Before making your second file edit, STOP. Count how many files you plan to touch total. If 3 or more → abandon inline editing immediately, write a TOML spec, di