Muse Skills

Uses

Deal Hunter watches marketplace listings for one narrow signal: a brand-new, private-seller listing priced at least half off its conservative resale value, with at least $40 of upside. It ignores everything else, including listings from dealers or store accounts.

It runs on a schedule you set, checks only what posted since the last run, and estimates a conservative resale value from real sold comps, never from the average asking price. A listing survives the filter only if both the price bar and the dollar-upside bar clear.

When nothing clears the bar, the run stays quiet. You get a message only when there’s a real candidate to look at, so a scheduled check never turns into noise.

What it includes

• Watch config for category, keywords, marketplaces, and area
• New-listing and private-seller filtering
• Conservative comp estimation from sold data
• The 50%-of-value and $40-upside filter
• A short report format, sent only when a candidate survives

Example

Input: watch config for “road bike, private seller, 20 mile radius”, scheduled to run every morning at 8am.

Output, on a morning with one hit:

Listing: 2019 Trek Domane SL5, 56cm. [link]
Ask: $650
Conservative comp: $1,400 (lowest of 4 sold comps, same size/condition)
Upside: $750

On a morning with no hits: no message.

Uses

Marketplace Scout takes a watchlist of item names and checks Facebook Marketplace, Craigslist, and eBay for listings worth buying to resell. It estimates resale value from real eBay sold comps, not asking prices, and nets out marketplace fees and shipping before computing margin.

You maintain the watchlist: item names, an optional target condition, and an optional max price per item. Marketplace Scout searches all three sources for each item and keeps only listings with a margin of at least $25 after costs.

A candidate list groups by watchlist item and ranks by margin. When none of the day’s listings clear the bar, the run sends nothing.

What it includes

• A plain-text watchlist format: item, condition, max price
• Cross-platform search across Facebook Marketplace, Craigslist, eBay
• Resale estimate from eBay sold comps (median of at least 3)
• Fee and shipping deduction before margin
• A grouped, ranked report, sent only when a candidate survives

Example

Input watchlist:

item: Nintendo Switch OLED, condition: good, max price: 180
item: KitchenAid Artisan mixer, condition: any, max price: 120

Output on a run with hits:

Nintendo Switch OLED
Listing: OLED console, good condition. Facebook Marketplace. [link]
Price: $150
Est. resale: $230 (median of 5 sold comps)
Margin: $54

KitchenAid Artisan mixer
Listing: Artisan 5qt, tested working. Craigslist. [link]
Price: $80
Est. resale: $145 (median of 4 sold comps)
Margin: $52

A run with no qualifying listings sends no message.

Uses

Chief of Staff Briefing turns your calendar, your inbox, and a goal file you maintain into one morning brief. It reads all three, never edits them, and reports back in a fixed order: priorities, meetings with prep notes, inbox items needing a reply, and one nudge toward a goal that’s gone quiet.

You keep a plain goal file with one line per active goal. The brief reads it each run to pick the single nudge; you update the file yourself as goals change.

Run it each morning, or put it on a schedule. It only reports; it never drafts a reply or touches your calendar.

What it includes

• The goal-state file template and how to maintain it
• The full brief prompt, with its four fixed sections in order
• Rules for what to say when a section has nothing to report

Example

Goal file:

- Ship irisfile beta: waiting on QA sign-off, no update in 6 days
- Close Northwind renewal: proposal sent, awaiting reply

Output brief:

Today's three priorities
1. Reply to Northwind before their 2pm call, proposal has been open 4 days
2. Prep for the 10am Atlas sync, QA sign-off still pending
3. Confirm irisfile QA status, no update in 6 days

Meetings with prep notes
10:00 Atlas sync, with Dana and Priya. Prep: get QA's blocker list before the call
2:00 Northwind renewal call, with Marcus. Prep: bring the revised proposal PDF

Inbox items needing a reply
- Marcus (Northwind), "Re: renewal terms": needs a yes/no on the discount ask
- Dana (Atlas), "QA blockers": needs a decision on whether to slip the date

One goal nudge
Ship irisfile beta: ping Dana today for the specific QA blockers before the 10am sync.

Uses

Inbox Triage runs on your connected email account every weekday morning and sorts what came in overnight. It splits new mail into four buckets: reply, delegate, archive, and later. For anything that needs a reply, it drafts one in your voice from the thread history, and for anything that needs someone else, it drafts a short forward note.

Nothing goes out on its own. Every draft, reply or forward, waits for your approval. You can approve a draft as written, edit it first, or discard it. Archive and later labels apply only after you have seen the batch.

The default run time is weekdays at 07:30 local. The assistant confirms that time with you before creating the recurring job, and asks again if you want to change it later.

What it includes

• A cron-based recurring job, default weekdays 07:30 local
• Four-way classification: reply, delegate, archive, later
• Drafted replies for every message in the reply bucket
• A per-message approval step before any send

Example

Input: 6 new messages overnight, including a client question about a due date and a newsletter.

Output: “3 reply, 1 delegate, 2 archive. Draft 1 (to client@example.com, re: ‘Due date?’): ‘Hi Sam, the report ships Friday, no change from last week. Let me know if that still works on your end.’ Approve, edit, or discard?” The two archive items and the newsletter are listed by subject line only, with no draft attached.

Uses

This prompt pack turns a messy dump of meeting notes into a record you can act on. It pulls out decisions, action items with an owner and a deadline, and any question the meeting left open. It ends with a three-line summary you can forward on its own.

It runs once, on demand, whenever you paste notes into the chat. It never touches a calendar or a task tracker; you copy the output where it needs to go.

An action item with no named owner does not disappear or get a guessed name. It stays in the list, marked “No owner assigned”, so you catch it before the meeting’s decisions go untracked.

What it includes

• A fill-in prompt template with a [Paste raw meeting notes here] slot.
• A fixed output format: decisions, action items, open questions, three-line summary.
• A rule against inventing an owner, deadline, or decision the notes do not state.

Example

Input (excerpt):

Standup 9/8. Discussed the Q3 report delay. Sarah will pull the numbers by Friday.
Someone needs to email the client about the timeline. Still unclear if we're
using the old template or the new one.

Output (excerpt):

## Decisions
- The Q3 report is delayed.

## Action items
- Pull the Q3 numbers | Owner: Sarah | Deadline: Friday
- Email the client about the timeline | Owner: No owner assigned | Deadline: No deadline set

## Open questions
- Are we using the old report template or the new one?

## Summary
The Q3 report is delayed and Sarah owns the numbers by Friday.
No one is assigned to email the client about the new timeline.
The report template to use is still undecided.

Uses

This config turns the assistant’s feed prompt setting into a briefing you control. You name the topics to follow, the sources you trust, and the sources or angles to skip. You also set how long a briefing runs and which day it arrives.

It lives in one place: the feed prompt field in your assistant settings. The install prompt shows you the filled-in snippet and waits for your confirmation before it writes anything there.

A week with no genuine update on any of your topics sends nothing. The config states that rule directly, so a quiet week doesn’t turn into a briefing padded with old news.

What it includes

• A YAML-style snippet with topics_follow, sources_prefer, sources_skip, output_length, and delivery_day fields.
• A stated location: the assistant’s feed prompt setting.
• A confirm-before-write step, and a quiet-week rule that suppresses empty briefings.

Example

Filled-in config:

topics_follow:
  - "Federal Reserve rate decisions"
  - "Changes to Section 8 housing subsidy rules"
  - "Blackstone real estate moves"

sources_prefer:
  - "The Real Deal"
  - "Bisnow"

sources_skip:
  - "general stock market commentary"

output_length: "short: 3-5 bullets"
delivery_day: "Monday morning"

A week’s briefing:

- The Fed held rates steady this week, citing housing-sector softness (Bisnow).
- No Section 8 rule changes this week.
- Blackstone announced a $2B acquisition of a logistics REIT (The Real Deal).

Uses

Subscription Audit looks at your email receipts and a bank export file you provide, once a month, and builds a picture of what you are actually paying for on repeat. It groups charges by merchant, works out how often each one bills and what that costs you over a year, then ranks the list from most to least expensive.

Every merchant on the list gets a call: keep, review, or cut. A cut or review call always comes with a specific reason, such as no matching usage in your receipts, a duplicate plan, or a price increase with no explanation. Nothing gets cancelled. You get the list and the reasoning, and you decide.

This workflow never logs into your bank. It reads your connected email inbox for receipts, and it needs you to export and hand over the bank or card transaction file each run. The default schedule is monthly, the 1st at 08:00 local, confirmed with you before the recurring job is created.

What it includes

• A monthly cron job, default the 1st at 08:00 local
• A merchant-grouping pass over email receipts plus your bank export
• An annualized cost estimate per merchant
• A ranked cut list, each entry with one concrete reason

Example

Input: email receipts plus a bank export covering the last 30 days, containing charges from Netflix, a gym, and a note-taking app used twice all year.

Output: “Estimated annual recurring spend: $612. Netflix: monthly, $15.49, $186/year, keep. Gym: monthly, $60, $720/year… review, no check-in activity found in receipts for 3 months. NoteApp: annual, $84/year, cut, only 2 sessions logged since last renewal.”

Uses

Gift Finder runs a short interview, then hands back five real gift ideas worth checking out.

It asks five questions: relationship, interests, budget, occasion, and what the recipient already has. From your answers it builds a shortlist of five gifts, spread across your budget, each with a one-line reason tied to what you told it and a link to the actual product page. It never buys anything. You decide, and you check out yourself, on the site you trust.

It needs honest answers to the five questions. A vague interest (“likes stuff”) gets a vaguer shortlist.

What it includes

• A five-question interview about the gift recipient
• A shortlist of five gifts, each with a reason and a verified buy link
• No purchase, no cart, no stored payment or shipping details

Example

Input: “Gift ideas for my sister. She’s into hiking and pottery, budget $60, it’s her birthday, and she already has a good day pack.”

Output:

1. Ceramic glazing starter kit: builds on her pottery hobby without duplicating gear she owns: [link]
2. Merino wool hiking socks (2-pack): useful on every hike, well under budget: [link]
3. Collapsible hiking mug: fits the hiking interest, leaves room in the $60 budget for a card: [link]
4. Pottery throwing tool set: a step up from basic tools for someone already into the craft: [link]
5. Trail map subscription (1 year): supports hiking without adding more gear to a pack she already has: [link]

Uses

Price Drop Watcher checks a list of product URLs once a day and compares each one’s current price to the target you set. When a price drops to or below target, you get one alert covering every product that dropped, with its current price and a link. When nothing dropped, you hear nothing. No daily “still no change” noise.

You build the list yourself. Give it a product URL and a target price to start tracking it (“track this at $40”). Name a URL or product to stop tracking it. Ask for the full list at any time to see every tracked item with its target and last known price.

The default check time is daily at 09:00 local. The assistant confirms that time with you before creating the recurring job.

What it includes

• A daily cron job, default 09:00 local
• A watch list with target price per product, editable anytime
• Add and remove commands for the watch list
• One alert per run, sent only when a price actually drops below target

Example

Input: watch list has a $120 jacket tracked at target $90, and a $45 blender tracked at target $40.

Run 1 output: jacket is $95, blender is $42. Neither hits target. No message sent.

Run 2 output: jacket drops to $88. Alert: “Jacket: $88, target $90. https://example.com/jacket”. The blender stays out of the alert because it is still above its target.

Uses

Trip Planner turns six details about a trip into a day-by-day itinerary you can act on.

Give it your destination, dates, budget, travel style, and must-dos, and it returns a plan broken into morning, afternoon, and evening blocks, each with a cost estimate and a link to research the booking. It fits every must-do into a specific day, flags anything that needs advance booking, and totals the cost against your budget. It never books anything: every price is an estimate for you to verify before you pay for it.

It needs real dates and a real budget number to build a plan that fits. A vague budget (“cheap”) gets a vaguer cost estimate.

What it includes

• A day-by-day itinerary with time blocks and durations
• A cost estimate per activity, meal, and transit leg, plus a running total
• A research link for each paid activity or booking
• A flag for anything that needs advance booking, and how far ahead

Example

Input: “Destination: Lisbon. Dates: June 10 to June 14. Budget: $1200 total for 2 people. Travel style: relaxed, mid-range hotels, mostly walking. Must-dos: Belem Tower, a fado show, pastel de nata tasting. Travelers: 2 adults.”

Output: A 4-night plan. Day 1 morning: walk Alfama district (free); afternoon: Belem Tower and pastries at Pastéis de Belém (€15 for two, [link]); evening: fado dinner show in Alfama (€70 for two, book 1 week ahead, [link]). Days 2 to 4 follow the same structure. Total estimate: $1,050, $150 under budget. Note at the end: all prices are estimates, verify current rates before booking.

Uses

Vinyl Vault catalogs a vinyl collection from photos of album covers or a pasted list, and tracks what each record is worth.

Send a photo or a list, and it records the artist, title, pressing, and condition for each item. Ask for the collection’s value, and it pulls recent sold prices for each pressing and reports a total, with any record it could not price called out by name. Ask it to watch for something, and it keeps a wantlist that clears itself once you own the title.

It needs clear photos or accurate pasted details, and an honest condition grade, since the value estimate depends on both.

What it includes

• A running catalog of the collection with pressing and condition detail
• A wantlist that tracks titles you’re chasing
• A value report built from recent sold prices, run on request or monthly

Example

Input: “Add these: Blue Note Miles Davis Kind of Blue, 1959 mono pressing, Very Good Plus. Also add Radiohead Kid A to my wantlist.”

Output: “Added Miles Davis, Kind of Blue (1959 mono, VG+) to your collection. Added Radiohead, Kid A to your wantlist. Want a value estimate for the Miles Davis pressing now, or with your next monthly report?”

Uses

This skill reviews a diff the way a senior teammate would: it reads the surrounding code, not just the changed lines, before it says anything. It reports what breaks the program first, then what a security review would flag, and it never touches your files.

It runs when you ask for a review of a diff, a PR, or a set of changes. You get a ranked list of findings, each one anchored to a file and line, with the smallest patch that fixes it. You decide which patches to apply.

The skill never rewrites your code on its own. Every fix comes back as a proposed diff, so the change that lands in your repo is still the one you chose.

What it includes

• A SKILL.md with Purpose, Workflow, Output Contract, and Operating Rules sections.
• A fixed review order: correctness bugs, then security issues, each ranked by severity.
• A rule that blocks the skill from writing, editing, or staging any file.

Example

Input: a diff adding a new /api/refund endpoint that reads an amount from the request body and calls charge.refund(amount).

Output (excerpt):

1. api/refund.ts:14
   Failure scenario: A caller sends a negative amount. The endpoint passes it straight to
   charge.refund, which credits the caller's account instead of refunding it.
   Fix:
   --- a/api/refund.ts
   +++ b/api/refund.ts
   @@ -12,6 +12,9 @@
      const { amount } = req.body;
   +  if (typeof amount !== "number" || amount <= 0) {
   +    return res.status(400).json({ error: "amount must be a positive number" });
   +  }
      const result = await charge.refund(amount);

2. api/refund.ts:14
   Failure scenario: The endpoint has no check that the caller owns the charge being
   refunded. Any authenticated user can refund any charge ID.
   Fix: add an ownership check before the call to charge.refund (ties to the auth
   middleware already used in api/charges.ts:22).