AI Agent Workflow Collection

Query the local Apple Mail (Mail.app) database on macOS to list, search, count, or extract content from emails. Always snapshots the SQLite DB first, queries the copy read-only, and suggests cleanup when done. Use when the user asks to "read my apple mail", "find emails from X", "list mail from sender", "search my inbox", "extract from email bodies", or otherwise wants to analyze locally synced Mail.app messages. Do NOT use for sending mail, modifying messages, configuring Mail rules/accounts, Gmail API / IMAP fetch, or non-Apple-Mail clients (Outlook, Thunderbird, Spark).

Export Apple Mail conversation threads from a given sender into one markdown file per thread, with an incremental manifest so re-runs only write new or changed threads. Use when the user wants to archive, download, or back up all emails from a sender to local .md files, group a sender's mail into thread files, or re-sync an existing mail archive to pick up new messages. Reads the local Mail.app SQLite index read-only through a /tmp snapshot and parses .emlx bodies. Do NOT use for sending or modifying mail, configuring Mail accounts or rules, Gmail API or IMAP fetch, non-Apple-Mail clients (Outlook, Thunderbird, Spark), or extracting attachments (text bodies only).

Chained planning workflow, one pass from a raw request to a hardened, cost-routed plan. First it sharpens your request via the prompt-enhancer skill. Then it builds a thorough implementation plan with plan-mode rigor (explore, design, draft). Then it stress-tests the plan via the grill-me skill, a relentless interview that resolves each decision branch and revises the plan. Then it routes the refined plan through the op skill, assigning each task the cheapest capable model and mapping dependencies. It presents the final routed plan and, once you approve, lets op dispatch the subagents and verify. It closes with a recap of the original and enhanced prompts and what each stage did. Slash-only. Use when you type /better-plan and want a plan enhanced, hardened, and cost-routed in one go. Do NOT use for a quick one-off plan with no review, to only grill an existing plan, or to only route an existing plan.

Mark a folder as trusted in Claude Code by setting hasTrustDialogAccepted in ~/.claude.json, skipping the interactive "Do you trust the files in this folder?" prompt — useful when provisioning a fresh server or running Claude Code non-interactively. Use when the user asks to "trust this folder in Claude Code", "skip the trust dialog/prompt", "allow my home folder", "make /root trusted", "pre-trust a directory", or invokes /claude-allow-home. Do NOT use to change tool permissions, allowlists, env vars, hooks, or any other Claude Code setting (use update-config for those) — this only flips the per-directory trust flag.

Run an extremely strict structural and architectural quality review on a diff, branch, or file set, hunting for "code judo" moves that dramatically simplify the implementation — not correctness bugs or style nits.

Run a question, idea, or decision through a council of 5-7 named AI advisors who answer independently, peer-review each other, then a chairman synthesizes one verdict. Slash-only — invoke it with /council-v2, it never auto-fires. Five permanent seats — First Principles (delegates to the first-principles-mode skill), The Contrarian, The Outsider, The Expansionist, and The Founder (delegates to the founder-thinking-mode skill). Two routed seats join only when the question fits — Stanier (delegates to persona-stanier) for people, teams, orgs, hiring, coaching, 1:1s, performance, OKRs, delegation, or engineering process, and levelsio (delegates to persona-levelsio) for solo or bootstrapped products, side projects, shipping, launching, pricing, distribution, or building in public. Do NOT use for factual lookups, questions with one right answer, trivial decisions, or anything without a real tradeoff and real stakes.

Generate an engaging STARTHERE.md codebase guide that explains architecture, decisions, and lessons in plain language with Mermaid diagrams. Also wires up auto-update checks in the project's agent instructions file and links from README.md. Use when onboarding to a project, documenting a codebase, or when user says "create codebase docs", "write STARTHERE", "explain the project", or "document the codebase".

Create a new implementation plan file for new features, refactoring existing code or upgrading packages, design, architecture or infrastructure.

Turn a short product or project description into a motivating vision doc covering three angles — motivation (the why and the shift it creates), practical (what using it actually looks like), and product (what it is and what to expect). The tagline ships in three wordings, a motivational main plus practical and product-descriptive alternatives. Use whenever the user wants a vision, mission, or why-this-matters framing for a product, project, tool, or workflow. Triggers — 'write a vision', 'draft a vision for X', 'give this a vision statement', or when the user pastes a short description of something they are building. Produces one tight vision doc (tagline wordings, what it is, what it does, what you get, the shift, success signal), not marketing copy or a pitch deck. Do NOT use to summarise existing text (use summarise-url or summarise-text), to break a goal into tasks (use goal-breakdown), or to plan an MVP launch (use ship-v1).

Generate production-quality SVG images (banners, cards, heroes, badges, posters) from a text description. Optionally enriches content by fetching a URL to extract branding, copy, and stats. Asks targeted clarifying questions one-at-a-time for missing info, then outputs a hand-crafted SVG file. Use when the user asks to "create an SVG", "generate a banner image", "make a card image", "create an OG image", "design a badge", or needs a vector image created from a description. Do NOT use for raster images (PNG/JPG), photo editing, or complex illustrations with many detailed shapes.

Distill raw notes into a sharp set of standalone maxims — distillation, not summarization. Keeps the vital few, drops 40-60% of ideas, compresses survivors to maxims of 8 words or fewer, promotes the most foundational idea to a headline, and sharpens contrasts into antithesis ("Give problems, not answers") or couplets ("Fewer tasks, bigger impact"). Returns flat, loosely clustered bullets in chat, then asks whether to also save them to a new .md file. Use when the user says "distill these notes", "turn my notes into maxims", "compress this to principles", "boil this down to maxims", or wants raw notes reduced to a vital few transferable principles. Accepts pasted text, a local file path, or an Obsidian note reference. Do NOT use to summarize while keeping all the ideas, to highlight key takeaways inside a note in place, or to build a reusable advisor persona from interview transcripts.

Split mixed raw notes into two outputs, a lossless organized reference for the facts and a sharpened set of maxims for the heuristics. Keeps every number, date, rate, threshold, condition, and obligation verbatim, groups facts by inferred category (deadlines, amounts, deductions, obligations, records), surfaces deadlines and action items, then boils transferable rules of thumb down to maxims of 8 words or fewer. Accepts pasted text, a local file path, or an Obsidian note reference, prints both sections in chat, then asks whether to also save them to a new .md file. Use for notes that mix reference facts with judgment calls, for example tax, medical, or legal notes, meeting minutes, or research logs, or when the user says organize these notes, structure my notes, sort facts from principles, or make a clean reference. Do NOT use when the input is purely principles and you only want lossy maxims, to highlight takeaways inside a note in place, or to build a reusable advisor persona.

Distill a leader's worldview from their interview transcripts into a reusable advisor persona. Produces a principles markdown (principles, mental models, recurring patterns, verbatim quotes, grouped by domain), a role markdown (description, core questions, mental models, tone), and a paste-ready CLAUDE.md activation snippet — all from the five-step "Distilling Leadership Wisdom" methodology. Asks one question at a time. Use when the user says "distill a persona from these transcripts", "build a leader's advisor persona", "extract leadership principles from interviews", "turn these podcast transcripts into a CLAUDE persona", or "create an advisor role from a leader's talks". Do NOT use for summarising a single article or note (use summarise-text or summarise-url), improving a prompt (prompt-enhancer), customer or user personas for product work (this skill is for leadership advisor personas only), or fetching transcripts (skill takes already-gathered local files, not URLs).

Translate any english text to czech version of it, while maintaining its accuracy. Use only when you are asked to translate to czech.

Strip a problem back to fundamental truths, question every assumption, and rebuild the answer from only what can be verified — instead of giving the conventional answer. Use when the user says "first principles", "enter first principles mode", "reason from first principles", "strip it back to what's actually true", "question the assumptions", "why is this really true", "don't give me the conventional/textbook answer", or wants received wisdom on a decision, belief, design, or estimate challenged from the ground up. Do NOT use for routine factual lookups, simple how-to questions, code execution, or when the user just wants a fast conventional answer.

Switch into a blunt, first-principles operator voice that gives the specific decision a seasoned founder who has built and exited companies would actually make — not balanced or generic advice. Every answer opens with the line 'Here's what I'd actually do', then names the call, the trade-off, the real risk, and what most people miss. Use when the user wants a direct verdict on a startup, product, indie-hacker, pricing, hiring, fundraising-vs-bootstrap, go-to-market, or pivot decision, or says 'founder thinking mode', 'founder mode', 'what would a founder do', or 'give it to me straight'. Do NOT use for routine coding or debugging, factual lookups, analysis the user explicitly wants balanced and neutral, or sensitive medical, legal, or compliance advice.

Generates a detailed, step-by-step developer task list in Markdown format from a Product Requirements Document (PRD). Use when you have a PRD and need to break it down into high-level parent tasks and granular sub-tasks for implementation. Triggers on requests like "generate tasks from PRD", "create task list for this PRD", or when an implementation plan is needed from a PRD.

Break a big finite goal, project, or idea into a sharp end state, ordered milestones, and one-day tasks with a single clear next action. Use whenever the user wants to decompose, break down, or plan a project into actionable steps, asks 'where do I start' on something they want to ship, build, learn, or launch, says 'make a plan to achieve X', 'turn this goal into tasks', '1% better every day toward X', or feels stuck because a goal is too big to begin. Also use to continue a plan already in progress: when the user finished a milestone and asks what is next, says 'continue', 're-plan from here', or 'update the plan', re-decompose the remaining work from where they actually landed. Handles finite projects with an end state only. Do NOT use for ongoing habits or systems with no finish line (there is no streak or habit tracking here), for pure scheduling or calendar work, for summarising web pages or text, or for distilling raw notes into maxims (a different job).

Highlight the most important takeaways and key learnings inside an Obsidian note by wrapping them in ==text== (Obsidian highlight syntax). Edits the note in place. Use when the user asks to "highlight key takeaways", "highlight key learnings", "mark the important parts", "find and highlight main points in ", or similar. Requires a note/file name as input.

End-of-session ritual that mines the current conversation for X/Twitter-worthy takeaways and drafts a build-in-public post. Runs directly on invocation — no permission question — reading the session on two lenses (the product you built plus the craft behind it), scoring candidates against an absolute resonance bar, filtering out anything employer-confidential or personally identifying, and refusing to force a post when nothing clears the bar. Surfaces a ranked shortlist, then drafts the chosen angle as a copy-paste-ready post. Tracks angles it has already pitched so it repeats one only when a new session is clearly stronger evidence. Use when you type /indie-hacker-wrapup, or say wrap up this session, draft a learning from this session for X, or session takeaways for twitter. Do NOT use for general prose writing, for summarising the session into notes, or to invent posts not grounded in what actually happened.

Audit landing page copy against a 13-section conversion blueprint and return a scored gap report + prioritized fix plan. Use when the user asks to "audit my landing page", "review LP copy", "gap-analyze a landing page", "score this landing page", "what's missing on this page", "compare my page to best practices", or pastes landing page text and asks for feedback. Accepts pasted markdown/text, a local file path, OR a URL (delegates URL fetch to the defuddle skill).

Audit a landing page against a fixed viral checklist, one verdict per check, then grill the gaps and plan the fixes. Scores every bullet in the checklist file in file order with verbatim quoted evidence, never inventing, merging, or skipping a check. Page-decidable checks are scored; checks about pricing model, positioning, or founder visibility are never scored from markup, they are deferred to a grill-me interview. Emits a gap report, then a remediation plan covering confirmed gaps only. Trigger phrases - "audit my landing page for virality", "run the viral checklist", "why won't this page get shared", "grill my landing page", "check this page against the viral principles". Do NOT use for conversion-rate optimization, SEO audits, accessibility review, or writing new copy. No composite score is ever computed.

Implement Microsoft Clarity analytics tracking in Next.js applications. Use when user wants to add Microsoft Clarity, set up heatmaps and session recordings, add analytics tracking to a Next.js app, or integrate Clarity for user behavior insights.

Implements Google Analytics 4 tracking with GDPR-compliant Silktide cookie consent in Next.js projects. Use when the user wants to add GA tracking, implement Google Analytics, set up analytics with cookie consent, or add GDPR-compliant tracking to a Next.js app.