coroboros

Systematic implementation using APEX methodology (Analyze-Plan-Execute-eXamine) with parallel subagents and self-validation. Use when implementing features, fixing bugs, or making code changes that benefit from structured workflow.

Use this skill whenever a user has an audio file (.wav/.mp3/.flac/etc.) that needs to loop as background on a website or web page — hero ambience, landing-page atmosphere, portfolio mood audio, ambient wind/rain/ocean/forest/breeze/pad beds, or any "play quietly behind the page while users read/scroll/interact" use case. Trigger on intents like making a clip loop without audible gaps or boundary artifacts on a site, fixing a click/tick/pop/bump at the loop boundary, fixing stereo bias or L/R imbalance on a web-bound ambient clip, normalizing loudness (LUFS) for web delivery, encoding a loop for web playback, or fading audio in on first user click/gesture — in any web framework (Next.js, Vue, Astro, Nuxt, plain HTML). The skill outputs a gapless FLAC plus a paste-in Web Audio snippet that unlocks on first interaction. Skip for general audio editing (cuts, mixing, effects), music/podcast mastering, or transcription.

Govern the BRAND-VOICE.md — extract a brand's writing voice from URL, Notion page, MD file/directory, or interview into a structured executable doc; update incrementally; diff versions; validate the canonical format; show testable rules. Supports multi-voice via `voice.extends` (founder on corporate, persona on institutional, multi-host) with `_replace`/`_remove` overrides. Consumed by writing skills via `-f` (e.g. `humanize-en`). Use when defining, extracting, refreshing, validating, or inspecting a brand's writing voice — "create a brand voice doc", "extract voice from this site", "tone of voice", "writing style guide", "BRAND-VOICE.md", "founder voice", "persona voice", "multi-voice".

Eight-axis judgment code review for the current diff — Correctness, Simplification, Tests, Documentation, Style, Intent, Design/API, Performance (+ Coherence on metadata changes). Five-phase pipeline scope → deterministic tool battery (npx/uvx-preferred, zero-install for the JS + Python majority) → 8 parallel LLM axis reviewers → Haiku validators on sub-80 findings (verbatim rubric, ≥80 threshold) → synthesis with no-silent-drop + Conventional Comments JSONL. Every report closes with "What I did NOT check" (security → /security-review, runtime perf, flaky detection). Opt-in flags `--verify-build`, `--mutation-test`, `--reconcile`, `--apply-safe`. Public-skill posture — zero auto-install, graceful skip on missing native tools. Invoke before a commit or PR — "review my changes", "deep review", "did I miss anything", "check before I commit", "audit this PR", "drift / gaps / blind spots".

Govern an existing DESIGN.md — Google's open standard for design tokens (YAML frontmatter + eight prose sections). Auto-activates during UI edits to enforce token-only sourcing for colors, typography, spacing, and corner radius when a DESIGN.md is present; steps aside when none exists — it never forces or authors one (`/award-design` authors it up front, then builds under it). Exposes seven CLI-backed subcommands — audit (lint + fix), diff (regression check), export (Tailwind / DTCG), spec (canonical spec), migrate (port legacy Stitch), init (minimal scaffold fallback), audit-extensions (drift check vs globals.css @theme). Token-affecting UI changes update DESIGN.md first, audit, then propagate to code.

Download video or audio from ~1800 sites (YouTube, Vimeo, SoundCloud, Twitch, X…) with the `yt-dlp` CLI — full videos, mp3 audio, playlists, time-range clips, subtitles, capped resolutions. Use whenever the user wants to download, save, grab, or rip a video/audio/playlist from a URL, extract a song as mp3, clip a segment, or fetch subtitles — even when they just say "download this", "get the audio", or "save this playlist".

Pre-implementation thinking — research the problem space, weigh approaches with devil's-advocate rigor, decide what's reversible and conventional while surfacing the load-bearing forks for the user, and emit one apex-ready plan. Use whenever a non-trivial task needs thinking, comparison, or decomposition before code — even when the user doesn't say "forge" (e.g. "should we", "what's the best way", "compare", "evaluate", "pros and cons", "think through", "plan this", "break this down", "spec out", "create issues for", "map out the steps"). Produces a Decision by default (terminal — discuss-then-build); promotes to a Spec with prioritized workstreams when the build path is clear. NOT implementation (→ /ultrapex, /apex, or /oneshot).

Convert any document to Markdown with Microsoft's `markitdown` CLI — PDF, Word, Excel, PowerPoint, HTML, CSV, JSON, XML, ZIP, EPub, images (OCR/EXIF), audio (transcription), and YouTube URLs. Use whenever the user wants to extract text from a binary document, transcribe audio, OCR an image, scrape a YouTube transcript, or pre-process a file for an LLM context window — even when they just say "convert this pdf", "what's in this docx", "transcribe this mp3", or "get the text out of this".

Vet an untrusted agent skill, plugin, or MCP server before it is installed or trusted. Runs the skillward CLI — the complete deterministic scanner ensemble, offline, fused into one verdict — then triages the report into a plain-language risk read with an install / don't-install / remediate call. Use this whenever someone is about to add or run third-party agent code (npx skills add, a marketplace or plugin, an https Git URL, a directory of skills, or a local folder), even when they only ask "is this safe?" rather than naming a scan. Trigger phrases include scan this skill, vet before install, audit this agent skill, is this skill or MCP server safe, check this repo before I trust it.

Consult a local mirror of Anthropic's official docs for authoritative answers about Claude Code (CLI flags, non-interactive mode, hooks, MCP, subagents, slash commands, skills, plugins, CLAUDE.md, sandboxing, worktrees, checkpointing, statusline, settings.json, agent teams), the Claude Agent SDK (agents, tools, state, streaming, managed agents), and the Claude API (messages, tool use, prompt caching, files, batch, citations, extended thinking, vision, pricing, rate limits, endpoints). Use whenever the user asks how anything in Claude's ecosystem works or behaves — features, flags, parameters, limits, configuration — including casual phrasing, French, typos, and questions that don't explicitly say "check docs". Also triggers on "does X support Y / can I do Z / what's the diff between A and B / how do I configure …" and on complaints about Claude Code behavior the user wants to change. Deterministic Grep + full-file Read over the local mirror — never training data. Cites paths and source URLs.