← ClaudeAtlas

principle-scope-fencelisted

Apply when executing against an approved plan, structure, or anchored change. The artifact authorizes exactly the change it names; work outside the fence is documented, never done silently.
bostonaholic/team · ★ 11 · AI & Automation · score 73
Install: claude install-skill bostonaholic/team
# Scope Fence The approved upstream artifact bounds the work: it authorizes exactly the change it names. Work outside the fence is documented, never done — and scope expands by changing the artifact, never by quietly exceeding it. **Why:** What was reviewed is the plan; silent expansion ships unreviewed work under a reviewed label. The fence is also what makes "done" decidable: a bounded change can be verified, an elastic one cannot. **Pattern:** - Do not add steps, slices, or features beyond the plan. A missing piece is documented as a finding, not implemented on the spot. - Refactor or "improve" adjacent code where the plan calls for it, and note the opportunity where it does not. - An applied fix stays bounded to the anchored file and lines it was approved for; a change that wants to grow returns for approval. - If scope must genuinely expand, update the governing artifact — and for a material change, go back through its review. - Never expand or shrink scope in silence: record the expansion, or the omission, loudly where the reader will look (`skills/principle-skip-loudly/SKILL.md`).