Stats
Actions
Tags
From supersaiyan
USE after interrogate-codebase. Drives the real UI via Playwright MCP through the read-only ui-validator subagent to sample normal/edge/failed cases. Code reading, debug modals, log dumps are NEVER substitutes. If Playwright or the UI aren't reachable, the agent makes them so autonomously before escalating BLOCKED. Skip only for pure dev tooling and zero-behavior refactors.
How this skill is triggered — by the user, by Claude, or both
Slash command
/supersaiyan:investigate-ui-firstThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Before writing or editing code for any user change, ground the work in
Before writing or editing code for any user change, ground the work in real UI observations — backend changes count too if they eventually surface on a screen.
Make Playwright reachable — autonomously. Invoke
playwright-mcp-setup. If the UI isn't running, start the dev
server, install deps, or find a navigable URL yourself. Only
escalate BLOCKED after exhausting self-resolution.
Identify the UI surface(s) from the loaded code.
interrogate-codebase has already put routes, pages, and
components in context — name the relevant surface(s) directly from
there. Only ask the user if the code is genuinely ambiguous.
Build a sampling plan. State it in one sentence before dispatching: 2-3 normal cases + 2-3 edge cases + 1-2 failed/empty cases. If the user reported a specific bug, include the bug context in the brief so the sub-agent verifies it as part of the sample — but the sampling stays broad, the goal is to catch adjacent issues alongside (or instead of) the reported one.
Dispatch ui-validator with: the user's request verbatim, the
surfaces, the sampling plan, and any reported-bug context.
Read the report. Reject any report without browser_evidence.
EXPLORED → present concrete findings to the user with case IDs
(including the reported bug's reproduction status if applicable),
confirm priorities, then codePARTIAL-ACCESS / NO-EXAMPLES-FOUND → ask only for the missing
access or seed dataBLOCKED → resolve the blocker firstLog unrelated bugs with log-unrelated-bug — one per finding.
interrogate-codebase: code grounding first so this
skill knows which surfaces to sample; UI grounding second.brainstorming: with N concrete UI
observations in hand, brainstorming becomes "given these findings,
what should we change?" — short and grounded.Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub florian60430/supersaiyan --plugin supersaiyan