Productivity

Drafts follow-up messages to prospects and clients in the freelancer's own voice, learned from 2 or 3 of their real messages. Use when the user asks to follow up, nudge, check in, chase a reply, chase a late invoice, or "write a follow-up to [client]". Drafts only; it never sends anything.

Walk someone through a weighted decision matrix to choose between real options with multiple competing factors — job offers, apartments, cities, vendors, degrees, cars, hires, tools, vacation spots, anything where "just pick the cheapest one" isn't good enough. Elicits the top 5-10 criteria that actually matter, weights them, researches how well each option satisfies each criterion (1-10, with a cited rationale per cell), computes weighted totals, and produces both a markdown table and a self-contained interactive HTML report where weights and included options can be adjusted live. Use this whenever someone is stuck comparing options, asks "which one should I pick" / "help me decide betwe...

Write a one-page brief before delegating work. Use whenever you start a subagent (Agent tool), a workflow, a background job or any long multi-step task for another Claude, and when you plan a large build in stages. The brief gives the agent files to read first, a goal, the owner's own words, measurable done criteria, machine and hard rules, a time budget, a test budget, a stop rule, priorities, swap-compatible deliverables, a progress file and a fixed return format. Also covers stage planning, critics without write access, a targeted test policy and measuring against a reference before quality work.

Structured Socratic tutoring when the user explicitly asks to be taught, quizzed, or drilled to mastery on a topic: diagnostic placement, a skill map with levels 0-5, spaced review, and a final assessment. Trigger phrases: 'teach me', 'tutor me', 'quiz me', 'drill me', 'help me master'. Not for one-off 'explain X' or 'what is X' questions.

Click-to-comment feedback on any web page (Agentation toolbar) that Claude reads back and applies. Two modes - local (you annotate your own dev app or any page, notes go straight to Claude) and shared (colleagues comment on a hosted site via a self-hosted notes server). The toolbar can go into an app, onto any HTML page via a script tag, or onto any site via a bookmarklet or the Chrome extension. Use when the user wants to annotate/comment on a page, leave visual feedback for Claude, set up "заметки/комменты на макете" for colleagues, or read, apply or restore that feedback.

Use when a user wants to practise salary, compensation, or job-offer negotiation with a recruiter, including full mock calls, focused bargaining drills, or different difficulty levels. Not for hiring-interview practice or handling a real negotiation on the user's behalf.

Operate skanvil, the CLI that manages libraries of Agent Skills (a skills/ folder of SKILL.md files, a README index and a .skanvilrc.json). Use it whenever you create, rename, re-categorise, delete, validate or index skills in a repository that has a .skanvilrc.json or depends on skanvil, when fixing skanvil validation errors, or when installing skills into a project's .claude/skills, .github/skills or .agents/skills folders.

Make a step-by-step launch checklist for a digital product - before launch, launch day, and the first week - including the exact clicks the seller must do themselves on their store. Use when someone says "help me launch my product", "what do I need before I publish", "launch plan for my template/guide/pack", or is about to put a product on sale for the first time.

读取一份已写好的 plan.md，把其中的任务拆给子代理，并为每个子代理确定角色、模型和思考强度（effort），输出一份新的分配文档 plan.dispatch.md，目标是在质量和效率基本不变的前提下节省额度。当用户提到"分配子代理""给 plan 选模型""省额度""subagent 分工""模型和思考强度怎么配"，或者给出 plan.md 并准备在 Claude Code 或 Codex 里多代理执行时使用。只负责生成分配文档，不执行计划本身。

Coordinate two AI development teams on one repository through explicit ownership, short status reports, merge trains, tree certificates, and owner decisions.

Write a complete, honest product listing (title options, one-line summary, description, what's inside, who it's for, FAQ, tags) for a digital product by reading the product's own files. Use when someone says "write my listing", "product description for my template/guide/pack", or is about to put a digital product on a store page.

Measure and cut the context an agent loads before any work starts (CLAUDE.md, AGENTS.md, GEMINI.md, rules, @imports, skill listings) and how large sessions grow, using local Claude Code and Codex logs; then slim instruction files and prove no line was lost. Use when sessions feel heavy, compaction or usage limits arrive early, an instruction file keeps growing, Claude Code and Codex behave differently in one repo, or for "context บวม", "ลด token", "CLAUDE.md ใหญ่". Not the live meter for the cur

A 10-minute privacy check to run before publishing any new or changed vibe-coded app, landing page, course site or funnel - ten pass/fail checks centered on "does anything track people before they click Accept?", the lesson from the tracking-pixel lawsuit video. Hands off to consent-audit and consent-gate when a check fails. Also a teachable lesson for students who build with Lovable, Bolt, v0 or Replit. Use when someone says "I'm about to publish", "ready to launch", "pre-launch check", "new Lo

Designs a short activity that checks one learning outcome, evaluates the learner's attempts with a rubric, gives targeted feedback and hints, records every attempt in SQLite, and decides whether the outcome is met. Use at the end of every teach-me step.

Fixes a website or vibe-coded app so no ad pixel, analytics, session replay or optional third-party script loads, fetches or sends anything until the visitor opts in - a default-deny consent loader, an honest banner (equal Accept and Reject, unchecked categories), withdrawal, Global Privacy Control, no customer emails or full URLs in tracking payloads, and an accurate privacy notice - then proves it with fresh-visit tests. Use after consent-audit finds tracking before consent, or when someone sa

Teaches one topic-map node in a one-on-one chat. It sets the node's expected outcomes, teaches them in small steps with worked examples, and ends every step with a test-me activity until all outcomes are met. Use after create-map, and whenever the learner continues learning.

Diagnoses what a learner actually wants to achieve when they say "I want to learn X", turns it into one observable, reachable goal, and lowers the goal when it is too far. Use at the start of any learning request, before check-me.

把 Obsidian 库变成 agent 的外部海马体——会话开始先读、结束必写、信息按 F0/F1/F2 三级置信度沉淀、每周回访遗忘。首次加载会在用户许可下建立 .AGENT/ 骨架与 AGENTS.md 规则。

Project-aware end-of-session closeout — refresh the read-first handoff memory + repoint MEMORY.md (works for ANY project), then sync the CURRENT project's state store if it has one. Use at session end, high context (~70%+), or "save context", "prepare handoff", "wrap up", "save the session". Auto-checkpoints mid-session.

Builds a personalized topic map, a small ordered prerequisite graph of nodes from the learner's current ability (from check-me) to their confirmed goal (from clarify-me), and saves it to SQLite. Use after check-me, or when a map needs nodes added.

Use when working with a project's roadmap, development plan, queue, work journal or item status: asking what is next, continuing or resuming planned work, recording a decision, updating an item, or reconciling status with evidence. Follows the project's own conventions. Triggers include «дорожная карта», «что дальше по плану?», «продолжай разработку по плану», «запиши решение», «обнови статус пункта».

Load whenever an agent has scheduled, uploaded or published content (posts, reels, stories, videos, files) on a social scheduler or a drive and must say what is actually live. Use it for "check what was actually published", "did it post", "is it scheduled", "nothing is published", "did the upload save", "revisa lo publicado", "¿ya salió?", "no salió en Instagram", "programado", "¿ya se guardó todo?", "nada está en el Drive". It builds a plan list, reads each item back from the scheduler's own li

Load before changing anything LIVE for a client or a business: pausing or turning on ads, campaigns or ad sets, changing budgets, turning off an app, bot, number or automation, cancelling a scheduled post, editing a live site, or any rule or script that can switch things off by itself. Use it for "only pause X", "just turn off Google Ads", "don't touch anything else", "solo le pausas", "solo eso", "no le apagues nada", "ni lo demás", "no toques lo demás", "sin tocar nada más", "que no apague nad

Emotional-discipline tool that makes the user plan for failure BEFORE a big money move. Runs a pre-mortem, stress-tests the position, writes kill criteria, checks exit costs and reversibility, separates story from numbers, applies a cooling-off gate, and saves a decision-journal entry. Also reviews past journal entries into a blind-spot report. Use whenever the user asks whether to commit money or take on debt, even if the question sounds routine, such as buying a stock, ETF, mutual fund, crypto, property or plot, locking savings in an FD or lock-in product, a personal or vehicle loan, EMIs, or buying on credit. Also use for FOMO, tips, limited-time pressure, pre-mortem, kill criteria, decision journal, exit plan, or review my past decisions, even if the skill is not named. Analysis and structure only. Never gives buy/sell advice. Not financial advice.