← ClaudeAtlas

loom-ralphlisted

Use when Loom work needs a bounded Ralph subagent, harness run, implementation slice, review pass, or worker handoff built from tickets, records, files, evidence, claims, or other bounded context.
z3z1ma/agent-loom · ★ 15 · AI & Automation · score 77
Install: claude install-skill z3z1ma/agent-loom
# loom-ralph Ralph is Loom's bounded worker and review discipline. A Ralph run has durable context in a ticket or owning surface and a thin transient launch transport. The durable context names the target, mission, read scope, write scope, stop conditions, evidence or review expectations, and where worker output will be reconciled. The launch prompt points at that context; it is not the durable contract. Ralph can use a harness subagent, headless harness command, manual handoff, or other transport. The consuming surface still owns the resulting state or judgment: tickets own execution and closure, evidence owns observations, audit owns adversarial review records, and other records keep their own truths. ## Use This Skill When Use this skill when preparing, launching, executing, or reconciling a bounded worker or review run from tickets, records, files, diffs, claims, evidence, or external references. Also use it when defining read scope, write scope, stop conditions, worker output, context style, or whether run context is current enough to launch. Do not launch Ralph to resolve unshaped product intent, policy, architecture, scope, data/state modeling, or design-coherence choices. Shape or route those first. ## Dispatch If preparing or launching: - read `references/run-shape.md` - read `references/running-ralph.md` - read `references/verification-posture.md` when implementation or validation evidence is expected - read records and source needed to bind the run - en