SuperAPI AI Directory

design-feedback-collector

Design & creation

View repository

Updated 10-16 02:28

About this skill

Design Feedback Collector

This skill turns comments into a controlled feedback register. It is not a visual critic by itself; it structures feedback so the chair and specialist skills can act without losing context.

Operating Modes

  • collect: gather user comments, agent critiques, and review-board notes.
  • triage: assign severity, owner, affected skill, affected subprocess, and decision status.
  • resolve: convert feedback into accepted, partially accepted, rejected, or applied decisions.
  • sync-dashboard: update review-board or task-plan-facing summaries.
  • sync-progress: update progress-screen user gates and feedback artifacts after a user decision.
  • handoff: send accepted feedback to the right specialist skill.

Required Workflow

  1. Read references/feedback-status-contract.md.
  2. Use assets/feedback-register.schema.json as the register shape.
  3. Separate feedback sources:
    • user comment;
    • colleague/agent critique;
    • validation finding;
    • council decision.
  4. Assign each item:
    • target skill or subprocess;
    • severity;
    • status;
    • owner;
    • required action;
    • evidence needed.
  5. Use color/status semantics:
    • blue: no comment/open empty state;
    • red: unresolved comment exists;
    • orange: partially accepted or partially applied;
    • green: accepted and applied.
  6. For progress-screen gates, preserve gate_id, selected option, comment, target phase/subprocess, and source artifact.
  7. Do not mark an item applied unless the changed artifact or decision record is referenced.

Safety Rules

  • Remove secrets, tokens, private URLs, .env values, recovery codes, and private screenshots from feedback before saving.
  • Do not collapse disagreement into a fake consensus; preserve rejected and partially accepted reasoning.
  • For high-impact disagreements, route to senior-ui-ux-orchestrator in council mode.

Handoff Shape

Feedback mode:
Target skill/subprocess:
Source:
Status:
Severity:
Decision:
Applied artifact:
Open risk:
Next owner:

Validation

When editing this skill, run:

python3 $CODEX_HOME/skills/.system/skill-creator/scripts/quick_validate.py $CODEX_HOME/skills/design-feedback-collector
python3 $CODEX_HOME/skills/senior-skill-architect/scripts/lint_production_skill.py $CODEX_HOME/skills/design-feedback-collector
python3 -m json.tool $CODEX_HOME/skills/design-feedback-collector/evals/evals.json >/dev/null