Open Design Review

Review a design

Use this skill only for an explicit request for the independent review and refinement cycle. The open-design skill owns the shared craft, project authority, and output contract. Ordinary design work stays with open-design-designer.

Run the existing definition

Inspect open-design-review through compozy__loop_inspect and read the live descriptors for loop operations. Use the active workspace; the open-design profile provides the designer, critic, and linter. If operations are unavailable, report that limitation without claiming an independent cycle ran.

Supply a concise brief and a workspace-relative artifact_path under docs/design/. Preserve existing or explicitly requested paths; otherwise use docs/design/<slug>/index.html. For related HTMLs, name the entry file and list the other boards or _uiux.md paths in the brief. The designer returns every HTML path for lint and review.

Pass the following config_overrides on both calls. CompozyOS runtime defaults override definition values, so the definition alone does not enforce this workflow's limit. full_body also reruns the complete workflow after an action failure. A rejected review always starts a fresh generation.

Dry-run with compozy__loop_run using dry: true, verify effective iteration_cap: 3 and reattempt_strategy: full_body, then execute with the same inputs and overrides without dry:

{
  "name": "open-design-review",
  "workspace": "<active-workspace-id>",
  "config_overrides": {
    "iteration_cap": 3,
    "reattempt_strategy": "full_body"
  },
  "inputs": {
    "artifact_path": "docs/design/session-actions/index.html",
    "brief": "Review and refine the existing session-actions board against the requested selection, confirmation, empty, and failure states. Preserve the product styling and scope."
  }
}

Optional designer and critic inputs select existing agent definitions. Provider and model selection follow CompozyOS's runtime settings. Do not create a new loop definition, specification, or registry to run this workflow.

Honor the bounded review contract

Each pass runs designer → native lint → independent critic → native lint verification. The designer refines the same files; the linter reads every returned HTML; the critic evaluates craft, purpose and states, brand, accessibility, and copy. The critic runs as a separate agent session with filesystem and browser tools, owns judgment, and returns a schema-validated verdict with evidence for every file. The designer owns corrections and receives the previous critic verdict, blockers, evidence, and exceptions on the next generation.

The linter must run successfully. Its findings are heuristic: inspect each one, require applicable fixes, and document a specific source-backed exception where appropriate. A lint passed value only reports its P0 result; it does not establish visual quality or complete review.

Completion requires an explicit approved verdict, no blocking issues, evidence for every reviewed file, and unchanged paths and lint digests before and after review. Each review identifies its file by lint_index, path, and digest. Its exceptions object must cover exactly the remaining lint IDs, with a specific source and reason for each; a clean file uses {}. IDs are unique within each artifact, and the lint remains the severity authority. The completion condition verifies this coverage and file identity without copying finding prose. The native completion condition fails on evaluation errors; missing or invalid critic output cannot become approval.