copilot-home-assistant

Execute pending agent-surface tasks by dispatching dedicated agents for each task. Quality-first approach — each task gets a focused agent, batched 3-4 in parallel. Use when user says "execute agent tasks", "run agent queue", "process agent-surface tasks", "agent task execution", or during the agent-task-executor cron cycle.

Hookflow governance philosophy and creation patterns — every behavioral correction becomes a deterministic hookflow rule. Use when: "create hook", "hookflow rule", "enforce behavior", "block pattern", "governance hook", "prevent mistake", "deterministic enforcement", "preToolUse deny", "postToolUse advisory", "hookflow audit", "new YAML hookflow", "behavioral correction", "hookflow-js"

{{PERSONAL_DOMAIN}} blog article creation — frontmatter schema, tag conventions, research standards, quality checklist, git worktree PR workflow, and video-companion article mode. Use when creating articles on {{PERSONAL_DOMAIN}}, writing blog posts, creating PRs on htek-dev-site, checking article quality, or any agent says "write article", "blog post", "{{PERSONAL_DOMAIN}} article", "create PR for article", "frontmatter schema", "article quality check".

Google Ads management via Playwright browser automation — navigation paths, campaign monitoring, conversion tracking, tag management, and performance analysis. Use when user says "check ads", "ad performance", "Google Ads", "campaign metrics", "conversion tracking", "ad budget", "check conversions", "Google tag", "ad management", "A/B test ads", "optimize ads", "ad spend", "bid management", "ad blocker dialog", or any Google Ads activity.

Vercel-connected site deployment workflow — branch + PR, wait for Vercel preview, extract preview URL, send to {{PARENT_1}} for review, merge only after approval. Use when agent says "create PR", "deploy preview", "Vercel preview", "review site changes", "push to Vercel", "deploy changes", "site update", "preview URL", "merge PR", or makes any code change to a Vercel-connected repository.

Content illustration workflow — generate the mandatory AI hero image first, then create contextual visuals (HTML→Playwright for simple diagrams, AI for concepts), wire heroImage/frontmatter, and inject inline visuals into MDX. Use when user says "illustrate", "add visuals", "hero image", "generate diagram", "content illustration", "architecture diagram", "process visual", "inline images", or "backpropagation visuals".

Parallel multi-agent dispatch and report compilation pattern — discover agents, launch in parallel, collect results, compile consolidated notification. Use when user says "orchestrate agents", "dispatch agents", "parallel launch", "compile reports", "agent check-in", "consolidated report", "multi-agent dispatch", or any orchestrator needs to coordinate multiple sub-agents.

End-to-end workflow for contributing fixes to the Copilot agent runtime ({{EMPLOYER_PARENT}}/copilot-agent-runtime): bug identification, fix implementation, PR creation, local patched build, installation, revert, and maintenance cron setup. Use when user says "copilot runtime", "patch copilot", "build copilot locally", "runtime contribution", "fix copilot bug", "copilot source", "build patched CLI", "runtime PR", "copilot-agent-runtime", or any Copilot CLI source-level contribution work.

AI image generation for {{{{EMPLOYER_PARENT}}_USERNAME}} social media — OpenAI gpt-image-2 API, {{PERSONAL_DOMAIN}} brand palette, professional infographic style, and upload workflow. Use when user says "generate image", "create image", "AI image", "post image", "infographic", "social media graphic", "image for post", or any image generation activity.

Central Time awareness and date computation for the {{FAMILY_NAME}} family platform — PowerShell CT time, relative date resolution, quiet hours, and time-based decision making. Use when user says "what time", "compute time", "quiet hours", "time awareness", "date calculation", "leave by time", "current time", or any time-sensitive operation.

Email scanning, categorization, and autonomous action pattern — read emails, categorize by type, take action (create tasks/events/bills), and send batched Telegram summary. Use when user says "scan emails", "email triage", "process inbox", "email check", "unread emails", "email actions", "handle emails", or any agent needs to read and act on emails.

Copilot CLI extension development and architecture — file structure, joinSession API, hook types, tool registration, env file patterns, polling intervals, and extension registry. Use when user says "create extension", "modify extension", "extension architecture", "how extensions work", "add tool", "extension hooks", "onPostToolUse", "onPreToolUse", "extension.mjs", or any extension development activity.

Safe file-writing pattern for large generated content — use create/edit and governed extension tools instead of huge PowerShell here-strings, Set-Content, Add-Content, Out-File, or shell redirection. Use when: "write article", "write post", "write agent instructions", "multi-line content", "PowerShell heredoc", "Set-Content", "Out-File", "large markdown", or "generated file"

Data governance pattern — protected files registry, extension tools for governed data, migration workflow from raw edits to validated tools. Use when user says "data governance", "protect file", "create data tool", "extension for data", "governed data", "protected files", "data migration", "schema enforcement", or any data persistence architecture activity.

Telegram messaging rules for the {{FAMILY_NAME}} family — TTS speak parameter, quiet hours, per-person formatting, and communication patterns. Use when user says "send telegram", "message {{PARENT_1}}", "message {{PARENT_2}}", "notify family", "telegram rules", or any Telegram communication activity.

VidPipe CLI video processing — run the VidPipe pipeline on a recorded video, monitor progress, launch the review server, and handle the full flow from recording to review-ready. Use when user says "process video", "vidpipe", "run vidpipe", "video processing", "review server", "pipeline log", "process recording", or when the video-bridge extension sends a processing request via session.send().

VidPipe record → process → review workflow for {{FAMILY_NAME}}-family — capture with video bridge, process with pinned VidPipe CLI, expose `vidpipe review` with `@ngrok/ngrok`, and notify {{PARENT_1}} via Telegram. Use when user says "VidPipe workflow", "process recorded video", "vidpipe review", "review link", "video bridge to VidPipe", or "record to review flow".

Client website domain agent lifecycle — shared development workflow, code quality, progress page maintenance, standup/check-in reporting, and diminishing returns detection for Astro + Tailwind + Vercel client sites. Use when building or maintaining any client website domain agent (client-site-template-a, client-site-template-b, etc.).

Brand protection rules for GitHub Copilot / Microsoft content — pre-publish review, competitor framing, and brand-safe content creation. Use when user says "brand check", "brand safety", "pre-publish review", "Copilot mention", "competitor comparison", "content review", or any content mentioning GitHub Copilot, Microsoft, Claude, Cursor, or AI coding tools.

Client proposal generation on {{PERSONAL_DOMAIN}} — create branded, password-gated proposal pages with phased scope, pricing, timelines, and shareable URLs. Use when user says "create proposal", "client proposal", "write proposal", "proposal for [name]", "new proposal", "send proposal", "build proposal page", "proposal draft", "update proposal", "revise proposal", or any client proposal activity.

Content analytics and engagement tracking for {{GITHUB_USERNAME}} — post performance, comment management, auto-replies, best posting times, follower growth, and strategy insights. Use when user says "check analytics", "post performance", "engagement stats", "comments", "reply to comments", "best time to post", "follower growth", "top posts", "content report", or wants performance data from Zernio/Late.

H-E-B grocery automation — catalog management, cart building, order history scraping, and delivery ordering using verified product data and Playwright scripts. Use when user says "grocery order", "H-E-B cart", "add to cart", "grocery list", "buy again", "weekly order", "remove from cart", "cart total", "what do we need from HEB", "build the cart", "grocery run", or wants to manage H-E-B delivery orders.

Lead management automation — per-lead folder creation, standardized templates, pipeline tracking, and data capture rules. Use when user says "got a lead", "new lead", "add lead", "review pipeline", "log call", "update lead", "lead pipeline", "lead status", "lead follow-up", "capture lead info", "onboard lead", "log communication", "lead research", "scope call notes", "lead pricing", "convert lead to project", or any lead/prospect management activity.

Nightly maintenance and reflection protocol — Phase 0 active maintenance (cron health, token health, task hygiene, budget sync, memory health) plus Phases 1-5 (session review, data gathering, reflection, proposals, report). Use when user says "nightly reflection", "nightly maintenance", "9 PM cron", "nightly report", "evening maintenance", "run reflection protocol", "maintenance cycle".

Calendly management via REST API v2 plus Playwright UI automation — event types, paid bookings with Stripe, availability, routing forms, webhooks, and authentication. Use when user says "Calendly", "booking link", "paid consult", "event type", "webhook", "PAT", "OAuth", "availability schedule", or "routing form".

Platform-specific social media content formatting — per-platform copy rules, hashtag strategy, voice guidelines, LinkedIn algorithm optimization, and engagement patterns for {{GITHUB_USERNAME}}. Use when user says "write post", "LinkedIn post", "social media copy", "platform formatting", "hashtag rules", "content voice", "engagement optimization", "write as {{PARENT_1}}", "social copy", or any multi-platform content creation activity.

Tool/MCP debugging escalation protocol — when a tool isn't working, stop fast, notify {{PARENT_1}}, and move on. Prevents context destruction and token waste. Use when agent says "tool not found", "MCP not available", "Playwright not working", "command failed", "trying to install", "debugging tool", or any repeated tool failure pattern.

Video publishing quality gate — Gemini-powered visual review, caption overflow detection, face visibility, audio quality, framing standards, and retry logic. Use when user says "review video quality", "quality check", "is this publishable", "check captions", "video review", "quality gate", "pre-publish check", or any video needs approval before publishing.

Cross-workspace communication with a work-agent repo via the agent mesh — calendar sync (personal→Outlook OOF blocks), work queries (pipeline, meetings), and cross-workspace task delegation. Use when user says "tell work agent", "sync to Outlook", "block Outlook", "ask work agent", "send to work", "work calendar", "OOF block", "mesh message", "cross-workspace", "relay to work", or any communication with the work-agent-repo workspace.

Safety-critical rules for child location tracking, pickup reminders, and caregiver handoff verification. Use when any agent mentions "child location", "{{CHILD_1_NAME}} is with", "babysitter", "{{CAREGIVER_NAME}}", "pickup time", "drop-off", "childcare", "where is {{CHILD_1_NAME}}", "caregiver", or handles any child-related logistics.

Engagement-respecting message cadence — anti-nag protocol, response tracking, escalating back-off, fatigue detection, and per-recipient adaptation. Use when agent says "check engagement", "skip cycle", "unanswered messages", "cadence check", "should I message", "engagement fatigue", "back-off", "nudge frequency", "message spacing", or any periodic messaging decision.

All food measurements must use grams exclusively. {{PARENT_1}} uses a kitchen scale — never use tablespoons, ounces, cups, or volumetric units. Use when providing recipes, meal suggestions, ingredient quantities, or any food-related measurements.

Inline task completion and serve-next pattern for the main session — handles "done/next" without launching full task-coach agent. Use when user says "done", "next", "finished", "move on", "completed", "what's next", "mark done", or any quick task transition that should be handled inline for speed.

Evidence-based recommendation protocol — search before advising, cite sources, cross-reference claims, and never hallucinate health/parenting/financial guidance. Use when user says "research this first", "cite sources", "evidence-based", "verify before suggesting", "don't make things up", "backed by research", or any agent needs to give health, medical, parenting, or safety advice.

Search and research tool priority hierarchy — which tools to use for web research, code lookup, and information gathering. Exa and Perplexity are PRIMARY. web_search/web_fetch are LAST RESORT. Use when any agent needs to search the web, research a topic, look up code examples, investigate a technology, or gather information from external sources. Trigger phrases: "search for", "look up", "research this", "find out", "investigate", "what is", "how does", "code example".

Post-shopping-trip workflow — check off purchased items, log expenses to budget tracker, and sync inventory. Use when user says "got groceries", "back from shopping", "bought the items", "went to store", "shopping done", "picked up groceries", "HEB run done", "Costco trip", or any indication a shopping trip was completed.

Team agent standup orchestration — phase-based sub-agent dispatch, standup prompt template, shared-agent dedup, progress collection, and consolidated reporting. Use when user says "team standup", "dispatch team", "team check-in", "run standup", "sub-agent dispatch", "team orchestration", or any team agent needs to coordinate its roster.

Twilio SMS messaging for external contacts — send text messages to service providers, clients, doctors, and non-family contacts. Use when user says "send SMS", "text message", "send text", "Twilio", "SMS to", "text {{EXTERNAL_CONTACT_NAME}}", "message provider", or any SMS/text messaging activity.

Work-hours task filtering and availability detection — suppresses physical tasks during meetings, serves only digital tasks during work blocks, and determines free gaps for task delivery. Use when checking availability, filtering tasks by work status, deciding what to serve during meetings, or any agent says "is he in meetings", "work hours", "suppress chores", "can I nudge now", "is he available", "free block", "meeting check".

Agent dispatch decision framework — when to launch new agents via task tool vs steer existing idle agents via write_agent. Use when user says "launch agent", "steer agent", "dispatch agent", "write_agent or task", "existing agent", "idle agent", "new vs existing", "agent decision", or any multi-agent orchestration decision.