| name | artifact-restraint |
| description | Enforce a closed content inventory for generic artifacts. Use when creating, editing, planning, or reviewing PDFs, resumes, documents, slides, roadshows, videos, tutorials, web pages, generative images, or similar deliverables. Apply for artifacts by default unless the user explicitly requests detailed or comprehensive treatment. |
Artifact restraint
Hard default
Treat user-supplied and verified content as a closed inventory. The artifact may
arrange and present that inventory; it must not enlarge it.
Every claim, action, benefit, example, section, label, object, and decoration
must map to one inventory item. State each item once. A repeated or unmapped
element makes the artifact invalid; revise before returning it.
Prevent content bloat, repeated or unsupported claims, narrative padding,
template-derived sections, and unsoliscated decorative entities.
Polished and professional change execution quality, not content scope. Format
conventions do not authorize content. A landing page does not inherently need
navigation, calls to action, feature sections, screenshots, links, or a footer.
Short output and empty space are valid.
Workflow
- State the artifact's purpose internally.
- Lock the inventory from user material and verified sources.
- Build the smallest structure that presents it. Use one content block when
one block is sufficient, especially for one or two facts.
- Audit every noun, verb, number, label, and visible object. Delete it if it
does not map to the inventory; delete later appearances of repeated items.
- Verify the domain claims that remain.
Omit missing optional information. Ask or use an explicit placeholder only when
the artifact requires the missing information to function; a conventional
template never creates that requirement.
An explicit request for detail or comprehensiveness expands only the named
dimension and never licenses unsupported facts. Keep mandatory accessibility,
safety, legal, and output-format requirements.
Apply this policy to artifact content, not implementation completeness, tests,
evidence, or ordinary collaboration. Let $writing-as-human govern requested
voice and $visual-qa verify rendered output.