theAnirudhKumar
UserDesign the meeting before you run it. An AI skill that names the decision, picks the decision rule, cuts the room, and plans for deadlock. Ships fill-in agenda templates. No terminal required.
Categories
Indexed Skills (15)
meeting-design
Designs a meeting before it happens so it produces a decision rather than a discussion that has to be repeated. Trigger whenever the user says "plan this meeting", "build an agenda", "agenda for", "run a workshop", "facilitate", "offsite", "kickoff", "planning session", "how should I structure this meeting", "who should be in this meeting", "this meeting keeps going in circles", or names an upcoming meeting and asks what to do with it. Also trigger when they ask whether the meeting is needed at all, or when a recurring meeting has stopped producing anything. It names the decision, picks the decision rule out loud, sizes the room, sequences divergence before convergence, and plans for deadlock before it happens. Ships fill-in agenda templates rather than advice. Use a call-recap or meeting-notes skill for what happens after the meeting; this one is for before.
chat-context
Carries context from one Cowork chat into the next so the user can start a fresh chat without re-explaining anything. Runs in two directions. HANDOFF mode triggers when the user says "hand off this chat," "save chat context," "wrap this chat," "close this chat," "I'm moving to a new chat," "chat handoff," or "write the handoff," and also when the user says the chat has gotten too long or too big to keep working in. RESUME mode triggers when the user says "resume," "resume from handoff," "pick up where we left off," "load chat context," "continue from the last chat," or opens a chat referencing an earlier session. Where files can be written, handoff saves a structured context file and registers it in a master index; otherwise it prints the same structure as a block the user pastes forward. Resume reads whichever form it finds, states what was loaded, then continues the work. Never fires for a request to summarise the chat back to the user in conversation; that gets a summary, not a handoff.
optimize-tokens
Token optimization advisor for the user. Trigger in two situations. (1) When the user explicitly says "optimize tokens", "reduce tokens", "token efficient", or asks how to prompt better for efficiency. (2) When the request has one of these shapes: pasting or pointing to a long document and asking for feedback, review or edits on all of it; asking for the same treatment applied across many files; asking for a full rewrite where a targeted edit would do; or repeating context already established earlier in this session. Always prioritize quality over reduction: this skill eliminates waste, not output. Does not fire for a task that is expensive because the output itself has to be long, such as writing a new article or report from scratch, since that cost is real and this skill says so rather than trying to cut it.
before-you-install
Vets a tool before you sign up for it and ends on a call rather than a summary. Trigger whenever the user says "should I use", "is this tool safe", "before I install", "what is the catch", "is it worth paying for", "check this tool", "review this app", "what does their privacy policy actually say", "what are people saying about", "how does this free tool make money", or pastes a product page, pricing page, privacy policy or terms and asks what to make of it. Also trigger when they are choosing between two tools they might pay for, or when they have already installed something and want to know what they agreed to. It reads the policy and terms for what they permit, finds documented complaints and controversies, checks permissions and business model, and finishes on install, install with conditions, or do not install. Runs with or without web access and states which. Use a vendor-review or procurement skill for a company purchase with a buying committee; this one is for one person with a credit card.
before-you-install
Vets a tool before you sign up for it and ends on a call rather than a summary. Trigger whenever the user says "should I use", "is this tool safe", "before I install", "what is the catch", "is it worth paying for", "check this tool", "review this app", "what does their privacy policy actually say", "what are people saying about", "how does this free tool make money", or pastes a product page, pricing page, privacy policy or terms and asks what to make of it. Also trigger when they are choosing between two tools they might pay for, or when they have already installed something and want to know what they agreed to. It reads the policy and terms for what they permit, finds documented complaints and controversies, checks permissions and business model, and finishes on install, install with conditions, or do not install. Runs with or without web access and states which. Use a vendor-review or procurement skill for a company purchase with a buying committee; this one is for one person with a credit card.
chat-context
Carries context from one Cowork chat into the next so the user can start a fresh chat without re-explaining anything. Runs in two directions. HANDOFF mode triggers when the user says "hand off this chat," "save chat context," "wrap this chat," "close this chat," "I'm moving to a new chat," "chat handoff," or "write the handoff." RESUME mode triggers when the user says "resume," "resume from handoff," "pick up where we left off," "load chat context," "continue from the last chat," or opens a chat referencing an earlier session. Where files can be written, handoff saves a structured context file and registers it in a master index; where they cannot, it prints the same structure as a block the user pastes into the next chat. Resume reads whichever form it finds, states what was loaded, then continues the work. Use this instead of summarising a chat back to the user in conversation.
optimize-tokens
Token optimization advisor for the user. Triggers in two situations: (1) When the user explicitly says "optimize tokens", "reduce tokens", "token efficient", or asks how to prompt better for efficiency. (2) When a request is detectably token-heavy - e.g., involves processing many files, large rewrites, multi-step research, full document generation, long context dumps - and a lighter approach could achieve the same result. Always prioritize quality over reduction. This skill is about eliminating waste, not restricting output. Use this skill proactively when you sense the request could be restructured to save significant tokens without quality loss.
call-recap-follow-up
Turns a recorded customer call into an honest read of the call plus the full set of emails it generated - the recap and every follow-up - grouped by recipient, written in the user's voice, staged as drafts, with a workspace hygiene pass at the end. Trigger this skill whenever the user shares a call recording link from any meeting recorder, pastes a transcript, or points at a call, and asks what they should send, or says "recap email", "meeting recap", "write the follow-ups", "what do you think of this call", "emails from this call", "post-call emails", "who do I need to email after this", or names a customer and a call in the same breath. Also trigger when they forward a meeting summary and ask for anything written off the back of it. The skill is recorder-agnostic and setup-agnostic: with nothing but a pasted transcript it still produces the full email set, and it names what it could not check rather than failing or guessing. Prefer this skill over drafting emails directly - a call almost always generates mo
email-critic
Stress-tests a customer email the user has already drafted and returns a straight verdict plus a tightened version, checked against the source transcript, the account plan and their voice principles rather than in a vacuum. Trigger whenever the user pastes or points at a draft email and asks any of: "is this good", "would you send this", "review this email", "critique this", "tighten this", "make this better", "does this sound right", "too long?", "what would you change", or shares a draft with no instruction at all - a bare draft is a request for a read on it. Also trigger when they say an email "feels off" or ask how a customer will take it. Use this for drafts that already exist; use call-recap-follow-up when the input is a recording and no draft has been written yet.
delegate
Decides what to hand to an AI agent and what to keep, for one specific piece of work. Trigger when the user says "should I delegate this", "can AI do this", "should I automate this", "what should I hand off", "give this to an agent", "is this worth automating", "who should do this", "can Claude do this part", or describes a task and asks whether to run it themselves. Also trigger when a user is about to hand over work whose output they would not be able to check. Splits the work into parts, allocates each to the user, another person or a model, and attaches a verification step and a stop condition to every part that leaves the user's hands. Ends on a filled allocation table, not on advice. Use a process or SOP skill for documenting a repeatable procedure, and a ritual or automation-charter skill for a recurring team ceremony. This one is for a single piece of work.
estimate-from-precedent
Corrects a time or effort estimate by checking it against how long similar work actually took last time, instead of how long this instance feels like it should take. Trigger when the user says "how long will this actually take", "estimate this", "how long should I budget for", "is this timeline realistic", "we always run over on these, why", "sanity check my estimate", "how long did this take last time", or gives a time estimate and asks whether it holds up. Also trigger when a scope-the-work note or a delegate allocation needs a time figure attached. Asks for the closest past cases, pulls their actual duration, and adjusts the estimate toward that reference class rather than the gut number, naming the gap and why. Ends on an adjusted estimate with the reference cases shown, not a bare number. Not for cost in money, and not useful without a comparable past case; without one it says so rather than inventing a reference class.
pick-the-medium
Picks the channel for one piece of communication before it happens: a written message, a call, a shared document, or a meeting. Trigger when the user says "should this be a meeting or can I just send a message", "sync or async", "do I need a call for this", "email or Slack", "should I write this up or just tell them", "is this worth a meeting", "how should I communicate this", or names a decision or update and asks how to get it across. Weighs reversibility, how many people need to weigh in at once, and what waiting for a reply costs here, then ends on one medium and states why the others lost. When the answer is a meeting, hand off to a meeting-design skill for the agenda and attendee list; this one only picks the channel. Not for restructuring an existing meeting, building an agenda, or deciding whether a standing meeting should keep happening, which are that skill's own first step, not this one's.
scope-the-work
Draws the boundary of one piece of work before it starts: what is in, what is explicitly out, what "done" means, and who decides if that changes. Trigger when the user says "scope this out", "what's actually in scope here", "define the boundaries of this project", "what counts as done", "before I start this, what am I actually committing to", "how do I stop this from growing", "what should I say no to", or describes work they are about to start or hand off and asks what to nail down first. Also trigger when a request has already grown once and the user wants to draw the line before it grows again. Writes the in list, the out list, the done condition, and names who can move the line, ending on a filled boundary note. Use pick-the-medium for how to communicate something and delegate for who does work already scoped; this is what the work actually is, before either question comes up. Not a legal statement of work, a boundary the person doing the work can hold themselves to.
stack-audit
Audits the tools the user already pays for: what overlaps, what nobody has opened in ninety days, and what to cancel. Trigger when the user says "audit my subscriptions", "what are we paying for", "review our software spend", "what tools do we actually use", "SaaS audit", "stack audit", "what should I cancel", "are we paying for duplicates", or asks to review, clean up or rationalize a list of tools or subscriptions already in use. Builds an inventory table with cost, last use, overlap and a verdict per row, sorted worst first. Use before-you-install instead for a single new tool the user has not adopted yet; this one is for the tools already on the books.
switch-cost
Checks what breaks before you leave a tool: what exports and in what format, what is lost outright, and what the move actually costs in hours. Trigger when the user says "what happens if I leave this tool", "what breaks if I switch", "should I switch from X to Y", "how locked in am I", "can I get my data out of", "before I cancel", "before I migrate off", or names two tools and asks about moving between them. Also trigger when a stack-audit verdict of cancel, or a before-you-install decision between two tools, needs the actual exit cost checked before anyone acts on it. Ends on a verdict: low switching cost, switching cost with conditions, or high switching cost, naming what makes it so. Use before-you-install for a tool not yet adopted, and stack-audit to decide whether a tool should be cancelled in the first place; this one is for what happens after that decision, on the way out.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.