| name | autodesign-visuals |
| description | Private visual reference subskill for prompt records, candidate records, and explicitly approved selected references. Does not create or fake images. |
Autodesign Visuals Contract
This private payload subskill is not public. Enter it only through autodesign-start or a later orchestrator after running the deterministic readiness check.
Required Upstream Artifacts
canonical.brand-direction at autodesign/outputs/canonical/brand-direction.jsoncanonical.screen-model at autodesign/outputs/canonical/screen-model.jsoncanonical.interaction-model at autodesign/outputs/canonical/interaction-model.json- generated
canonical.visual-anchor-proposals at autodesign/outputs/canonical/visual-anchor-proposals.json
Output Artifacts
visual.reference-prompts at autodesign/outputs/visual-references/prompt-records.jsonvisual.reference-candidates at autodesign/outputs/visual-references/candidate-records.jsonvisual.reference-selection at autodesign/outputs/visual-references/selected-references.jsonvisual.reference-set at autodesign/outputs/visual-references
Commands
Plan prompt records:
node autodesign-start/assets/payload/scripts/generate-visual-references.mjs --workspace <workspace> --action prompts --plan
Apply prompt records only after explicit approval:
node autodesign-start/assets/payload/scripts/generate-visual-references.mjs --workspace <workspace> --action prompts --apply --approve-visual-prompts --actor <actor> --at <timestamp>
Record candidates only after the active agent has generated real image files and canonical.visual-anchor-selection is manually approved:
node autodesign-start/assets/payload/scripts/generate-visual-references.mjs --workspace <workspace> --action candidates --prompt-id <prompt-id> --generated-output-path <path> --apply --approve-visual-candidates --actor <actor> --at <timestamp>
Select references only by explicit candidate id and explicit approval:
node autodesign-start/assets/payload/scripts/generate-visual-references.mjs --workspace <workspace> --action selection --reference-id <candidate-id> --apply --approve-visual-reference-selection --actor <actor> --at <timestamp>
Hard Gates
scripts/can-run-subskill.mjs --workspace <workspace> --subskill visuals must pass.canonical.visual-anchor-proposals must exist, be marked generated in the artifact graph, and have a manifest generation record.- Prompt records must include prompt text, an active-agent image-generation instruction, generated output path fields, and review metadata.
- Candidate recording must stop unless
canonical.visual-anchor-selection is approved by a manual gate record.
- Candidate recording must validate existing generated image output paths inside the workspace.
- Selected references must be supplied by
--reference-id and --approve-visual-reference-selection; the script must never auto-select.
Fail Fast
- Stop if state validation fails.
- Stop if graph dependencies are missing or cyclic.
- Stop if any required upstream artifact path is absent.
- Stop if canonical visual anchor proposals are not generated.
- Stop if candidate recording lacks manual anchor approval.
- Stop if a generated output path does not exist or is not an image file.
- Stop if selection lacks explicit reference ids and selected-reference approval.
The visuals subskill records prompts, candidate metadata, output paths, review metadata, and selected references. It must not create or fake images, perform Pencil operations, generate design-system artifacts, generate prototypes, or generate handoff material.
