Generate or edit bitmap images from Claude Code by delegating to Codex CLI image generation, without managing OpenAI API keys. Use when the user asks to generate or edit a raster image, or when an active task's spec explicitly designates codex-image as the image path (e.g. deck builds); never research raw codex invocation instead of loading this skill. When the grounding or scope is ambiguous, confirm before generating — image generation has cost and side effects.
Use when the user asks an in-depth question about a GitHub repository's internals, architecture, or implementation details — anything that benefits from AI-summarized repo documentation rather than raw code reading. Triggers on phrases like "how does <repo> handle X", "explain the architecture of <repo>", "what's inside <owner/repo>", or "compare <repoA> and <repoB>". Wraps the DeepWiki MCP (`mcp__deepwiki__read_wiki_structure` / `read_wiki_contents` / `ask_question`).
Onboard a full Playwright E2E test harness in the current project — verify/install Playwright, generate the official planner/generator/healer agents via npx playwright init-agents --loop=claude, scaffold auth separation (storageState + setup project), network route mocking, an E2E operating SSOT doc, and a gated GitHub Actions CI workflow with trace artifacts and PR-failure comments. Use when the user asks to set up E2E, add Playwright AI agents, bootstrap end-to-end testing, or wire E2E into CI. Never overwrites an existing playwright.config (merge proposal + backup). Degrades gracefully when Playwright is absent. Run from the user's project root, not this marketplace repo.
Use to scaffold the LLM-Wiki knowledge system (`.llmwiki/` insight + wiki + raw, plus spec) in a new repo that doesn't have it yet. Copies templates from the plugin's bundled assets, prompts for project pitch and 1-3 first domains, writes a slim CLAUDE.md.
Use after producing a new audit md, after merging a PR with non-obvious findings, or when a debugging session uncovers lore worth saving. Pulls the finding into the wiki with reversible diff-log + multi-page cross-update. Universal — works in any repo with a `.llmwiki/wiki/` or legacy `.claude/wiki/`.
Use when the user asks to audit wiki health, or periodically (manual trigger) to catch the 4 wiki-rot failure modes — identity duplication, level flattening, monotonic relationships, and staleness. Universal — works in any repo with a `.llmwiki/wiki/` or legacy `.claude/wiki/`.
Use to migrate or upgrade an existing wiki to the v2 `.llmwiki/` layout — triggers on "migrate wiki", "upgrade wiki to v2", "consolidate .claude/wiki and .codex/wiki", "move wiki to .llmwiki". Detects legacy `.claude/wiki/` and forked `.codex/wiki/`, consolidates them under the neutral `.llmwiki/` root, and adds v2 frontmatter. Universal — works in any repo with a legacy wiki.
Use as the verification gate BEFORE acting on remembered global/project guidance, provider quirks, design rationale, debugging stories, or module maps that aren't code invariants — check the wiki MOC first so you rely on the dated, sourced page instead of stale memory. Universal — works in any repo with a `.llmwiki/wiki/` or legacy `.claude/wiki/`.