Claude Skills by zalogarcia

List all /autopilot worktrees + branches in the current repo with their terminal_state, commits ahead of a base, files touched, and task summary. Use when the user asks which autopilot runs finished, are still pending, or are left to merge, or as the discovery step inside /autopilot-merge. Returns a TSV table.

Run Black Umbrella's own cold DM outreach. Triggers "run today's cold outreach", "cold DM batch", "outreach replies", "bu outreach", "approve the batch and send", "outreach report". Astra executes it in the codex-bare Codex session with Computer Use, sending from Zalo's real LinkedIn, Facebook and Instagram accounts in Chrome, to HVAC and plumbing owner operators countrywide (whatever metros prospects.csv holds; tranche 1 is twelve Sun Belt metros plus a nationwide LinkedIn rail). Enforces ON...

Crawl websites using Cloudflare Browser Rendering API. Use when the user wants to scrape, crawl, or extract content from a website (single page or multi-page). Returns HTML, Markdown, or structured JSON.

Create a multi-line git commit using the correct heredoc-inside-$(...) quoting pattern, with the Co-Authored-By trailer. Use whenever the user asks to commit, when a workflow needs to commit changes, or when a multi-line commit message is required. Eliminates re-deriving the tricky quoting and ensures a consistent commit-message style.

Finish a video Zalo recorded himself: rough cut the raw take first (fillers, dead air, false starts and superseded retakes out, cut on Scribe word boundaries and self checked at every cut), then add the headline, prompt/template cards, karaoke captions and hook-variant splits, then gate it against the locked safe-zone and caption laws. Use when he hands over self-recorded footage and describes overlays, captions, a headline or hook variants, or asks to cut the ums, the pauses or the retakes. ...

Author a new Claude Code skill in the established shape, or decide whether a recurring pattern deserves one. Use when the user wants to create, add or promote something to a skill, or asks whether it should be one. Covers the worth-it gate, the markdown-vs-script form factor, the SKILL.md template and the optional CLAUDE.md callout.

Run Zalo's daily content hour for the Zalo Kabche brand from ANY working directory — including Telegram/bridge sessions rooted at ~/dev. Invoke when Zalo says /daily, "what's today", "daily check-in", "today's reel", or opens a session to do his content hour. This is a thin router: the real procedure lives in ~/dev/zalo-kabche-brand/.claude/skills/daily/SKILL.md and is followed verbatim. Inside that repo, the repo-scoped skill wins and this one is unused.

Kill any stale dev server on a given port, restart it cleanly in the project, poll until it accepts connections, and smoke-test a route — returning HTTP status + log tail. Use before testing UI in the browser, after env changes, or whenever Claude wrote "pkill -f 'next dev'" in a previous turn and needs the server back up. Replaces the 100+ hand-written kill+sleep+curl variants.

Dubs the voice in a video or audio file into another language via the ElevenLabs Dubbing v1 API (English→Spanish by default, 90+ languages). Preserves each speaker's tone and pace; multi-speaker aware. Accepts a local file or a URL (YouTube, X, TikTok, Vimeo, direct link). Use when asked to "dub this video", "translate the voice/audio to Spanish", "make a Spanish version of this video", "localize this clip", or "revoice this in <language>". NOT for lip-sync (ElevenLabs replaces audio only — m...

Create distinctive, production-grade frontend interfaces with high design quality. Use this skill when the user asks to build web components, pages, or applications. Generates creative, polished code that avoids generic AI aesthetics.

Generate dense, text-heavy educational infographics via AI text-to-image — gpt-image-2, executed through the image-craft-expert agent. Encodes the 6 layout archetypes, 3 style presets, the quoted-string text-budget prompt architecture, and a mandatory read-back text audit. Use when the user says "make an infographic", "turn this into an infographic", "visualize this concept/framework", "visual cheat sheet", "diagram this idea", or hands over educational/marketing content to be turned into a s...

Run a designed live-test campaign against a deployed app or feature: code design review first, then a cheapest-first, destructive-last phase ladder with positive-evidence and state-neutralization rules. Use after shipping or before launch, or when the user asks to live test everything or catch all edge cases.

Generate branded motion-graphics "slides b-roll" for Zalo's videos (VSLs, YouTube, teleprompter scripts, reels) using the Machine Editorial v3 system in the Remotion studio at /Users/zalo/dev/operator-broll. Use when the user says "slides b-roll", "b-roll for this script", "motion graphics for my video", "VSL-style slides", "machine editorial", or hands over a script/teleprompter text and wants the graphics cut. NOT for the video game b-roll layer on a delivered reel (that's dot-matrix-broll)...

Scaffold a repo for Claude: scan the codebase, actually run the build/test/typecheck/lint commands, and generate .claude/CLAUDE.md, path-scoped .claude/rules/*.md, the machine-readable .claude/VERIFY.md verification manifest (deploy surfaces plus THE proof signal each deploy claim requires) and its runnable projection .claude/verify.sh. Use when the user asks to initialize or scaffold a repo for Claude, or when a repo lacks .claude/CLAUDE.md. Idempotent: fills gaps in existing scaffolds, neve...

Dump the Delta Agents production Supabase public schema (tables, columns, types, nullability, defaults, FKs, RLS flags) via the Supabase MCP and rewrite docs/SCHEMA-PROD.md with a generated-at header and drift notes. Use after applying any migration, whenever the snapshot is >7 days old, when a prod SQL query fails on a column/table name, or when the user asks to "refresh the schema snapshot". Prevents schema-guessing in ad-hoc SQL. The queries target Delta Agents prod ($DELTA_PROD_PROJECT_RE...

Generate a detailed, shot-by-shot, copy-and-paste-ready video prompt for Seedance 2.0 from a creative brief. Use whenever the user wants a Seedance prompt or a shot list, or describes a scene, ad concept, brand film, product video or other visual sequence they want turned into a generation-ready video prompt, even if they don't say "video prompt". Not for generating the clip itself (the seedance skill) or for branded motion-graphics slides (machine-editorial-broll).

Generate AI video clips (text, image or video to video, 4-15s, optional synced audio) through Segmind Seedance 2.0''s generate.sh. Use for any AI video generation request: a clip, b-roll, animating an image, a multi-shot sequence.

Ship delta-agents to production as one procedure — confirm target branch once, pre-flight via scripts/check.sh, heredoc commit, push, background-watch the "Deploy to ECS" run, ground-truth-verify per-service ECS image tags against the pushed SHA, smoke-test the gateway /health endpoint, and report. Use when the user says "ship it", "ship to prod", "deploy to prod", "push and deploy", or "commit, push and watch the deploy". Replaces the hand-rolled 200+-command commit/push/poll ritual with one...

End-to-end checklist for shipping a Zalo Kabche long-form YouTube video so no packaging or publish step gets skipped — edit-quality pass, a 3-TITLE split test pulled from the title-structures database, a 3-thumbnail split test (yt-thumbnail skill), a chapters-timed description, the correct upload path (Studio direct for long-form — Post for Me chokes on it), and the Studio upload, setup and publish through COMPUTER USE (computer-use.sh, Zalo's rule 2026-10-03). Use when the user says "ship th...

Send messages, files, images, or documents to the user via their Telegram bot. Use when the user asks to send something to Telegram, notify them, or share content via Telegram.

Evidence-first triage for "tenant X says Y broke, check why" reports on Delta Agents prod. Resolves the tenant by fuzzy slug match, time-scopes from the user's report, pulls parallel evidence (sessions, messages, voice calls + tool events, followups, alerts/audit/escalations) via the Supabase MCP, greps ECS gateway/worker logs by tenant id, correlates with recent deploys, and emits a triage summary BEFORE any hypothesis. Use for a tenant or customer complaint with a fuzzy symptom (a tenant, u...

Run TypeScript typecheck and production build for a Node/Next.js/Vite project, capture exit codes, and surface only the failure region (not the full log). Use when verifying changes before commit, in QA loops, after a refactor, or whenever a "did this build?" check is needed. Replaces hand-written tsc + build chains with a consistent invocation and consistent output shape.

Write Retell-style voice or chat agent system prompts (single bot, or orchestrator plus sub-agents) in the Operator Base house format, with the Selling Method block for sales bots. Use when the user wants a new client agent prompt or a multi-bot routing setup; not for small wording edits.

Triage one voice call (or "the last call") on Delta Agents prod — pull the tenant_voice_calls row (status/duration/disconnection reason), the tenant_voice_tool_events timeline, the gateway ECS log window around the call, and a DA-intended vs Retell-actual config diff via the Retell GET endpoints; for GPT-Live (engine openai-live) calls, the Twilio CallSid path instead: engine session, voice-bridge and gateway logs, the in-call booking guard. Use when a specific voice call misbehaved (hung up,...

Bring the full Delta Agents local voice-test stack into a known-good state — kill stale gateway/dashboard listeners, start both from a named checkout/worktree with nohup + readiness polling, start the tunnel (cloudflared or ngrok, auto-picked from GATEWAY_URL), and print the callable state (tunnel URL, gateway health, dashboard URL, webhook URL shapes). Use before any live voice/webhook test against the local gateway, when the user says "spin up the voice test stack", or whenever a Retell/CRM...

Generate Zalo Kabche YouTube thumbnails (1280×720) from the locked Shop Manual '74 spec using real reference photos of Zalo, via the image-craft-expert agent on the gpt-image-2 `images.edit` rail (OpenAI, the standing preference, NOT nano-banana), with a mandatory text + face-identity audit and the 120px squint test. Use when the user says "make the thumbnail", "thumbnail comps for video N", "thumbnail options", or at the packaging stage of any video in ~/dev/zalo-kabche-brand. Replaces one-l...

Make an explanation easy for Zalo to digest. Text always comes first, in STE-lite (80% of ASD-STE100, checked by ste-lint.py). At most ONE extra goes with it, chosen by pick-format.py and never by asking him. The extra is an exact boxes-and-arrows diagram PNG (flow, sequence, compare, map) or a self-contained interactive HTML page. Use when you explain a system, a report, a plan or a result to Zalo, or he asks to "break this down", "make this easier to digest", "diagram this" or "show it as a...

Produce new whiteboard YouTube video ideas for Zalo Kabche's channel in two stages. Stage 1 picks topics from the owner question research and his demo data (skipping anything filmed or planned on Linear) and sends them as ONE Telegram text with Linear Idea cards. Stage 2, only for the ones he picks, writes beats.md, renders a gpt-image-2 board per pick on the photo of his real whiteboard, builds the filming plan PDF ("like last time"), sends it with the boards and moves the cards to Boards re...