← ClaudeAtlas

persona-teardownlisted

Use when the user asks to tear down, walk through or review a prototype, UI kit, exported design or live screen as a persona or a panel of personas ("teardown", "rip it apart as X", "walk the prototype as the customer", "persona review"). Runs the persona-teardown SOP: cold persona agents driving agent-browser, exhaustive coverage ledger, one finding per line, consolidated into a design-tool amendment prompt.
Traction-AI-Labs/skills · ★ 1 · AI & Automation · score 74
Install: claude install-skill Traction-AI-Labs/skills
# Persona teardown Three to five persona agents independently walk every screen, control and state of an artefact in a real browser and report every defect one per line with a screenshot each. The orchestrator merges the reports mechanically into one amendment prompt for the design tool. No code changes; the agents never fix anything. Full reasoning, the source numbers behind every rule, the survey of existing skills and what was deliberately not adopted from them: `references/persona-teardown-sop.md`, bundled with this skill (the research note behind it is not bundled; the SOP runs without it). Read the SOP if a rule here looks arbitrary or the case needs adapting. Do not use this to learn what real users think. A persona is a search heuristic for finding defects, never evidence about people. Findings must be checkable from their own record. ## Orchestrator checklist Gather all of these before dispatching anyone, then dispatch all agents concurrently. - [ ] **Artefact served over HTTP**, URL written down. For a static export: `python3 -m http.server 8787 --directory <folder>`. Never `file://`; file-origin rules break fetch-based drawers and theme toggles. Load it once yourself first. - [ ] **Screen inventory** derived from the Screens drawer, nav or index, written to a file the agents read. The orchestrator may read the artefact's markup to derive it; the personas may not. - [ ] **Persona definitions**, one per agent (template below). Three is the def