Free AI skills for working musicians

---
name: facts-sheet
description: Build, update or audit a one-page facts sheet for a musician's or creative's business (offers, set lengths, service area, what's included, how to book, and the questions clients always ask). Use this whenever a musician wants to fix their bio, website copy, Google Business profile, EPK basics or FAQ, wants to be found and quoted correctly by search engines and AI assistants, or is about to draft replies to inquiries, even if they never say "facts sheet".
---

# Facts sheet

The facts sheet is the one page every reply, bio, FAQ and AI answer gets built from. An assistant, human or AI, can only be trusted to answer for you when the facts are written down. That makes this page the foundation for almost every other skill in this set.

## Two modes

- **Build:** there's no facts sheet yet. Interview, then write it.
- **Audit:** the user pastes their site copy, bio or profile. Compare it against the facts sheet (build one first if needed) and show what conflicts or is missing.

## Build: the interview

Ask one question at a time and wait for the answer. People answer one question far better than a form. Cover these, in order:

1. **Who you are, in one line:** instrument or act, style, city.
2. **What you offer:** each option with its name, length, lineup and what's included.
3. **Where you work:** service area, and whether you travel.
4. **How booking works:** the first step, what you need from the client, how a date gets held, and what makes it booked (for example, a signed agreement and a deposit).
5. **Logistics clients ask about:** setup time, space, power, sound, breaks.
6. **What you don't do.** This line prevents more bad replies than any other.
7. **The questions clients always ask.** Ask for the ten most common, then answer each one together.
8. **Proof you're allowed to name:** credits, venues or clients, only with their permission.
9. **Links:** site, booking link, listening links.

**Prices:** ask whether they want prices public. Many working musicians keep them off the site and send a price list after someone leaves an email or a phone number. Don't push either way, and never fill in a number they didn't give you.

## Output

Plain sentences under these headings, about 400 words at most. Mark any gap as **MISSING** instead of guessing.

```
# [Name]: facts sheet (updated [date])
## In one line
## What I offer
## Where I play
## How booking works
## Logistics
## What I don't do
## Questions clients ask
## Proof (named with permission)
## Links
```

## Audit: the comparison

Put each claim from their site, bio or profile in a table:

| Claim on the page | Facts sheet says | Status |
| --- | --- | --- |

Status is Match, Conflict or Missing. List the conflicts first, because those are the lines that cost bookings.

## Checks before you hand it over

- Every line is something the user told you, not something borrowed from a template or a competitor.
- Numbers (lengths, areas, lineup sizes) agree everywhere they appear.
- No client or venue is named without permission.
- It fits on one page.

## What to do with it next

Paste the same wording into the site, the Google profile and every bio, so people, search engines and AI assistants all read the same facts. The first-reply skill uses this sheet as its only source.

*From the Blues People AI Agency course, chapter 3: Your facts.*

---
name: first-reply
description: Draft the first reply to a gig, booking, lesson or collaboration inquiry in the musician's own voice, using only facts they have written down, so a lead hears back in minutes instead of days. Use whenever someone pastes an inquiry, DM, email or contact-form message and wants an answer, asks how to reply to leads faster, or wants auto-reply text for a form or CRM, even if they don't mention speed.
---

# First reply

Your website is not the business. Your response time is. The person asking about a date is often asking two or three other musicians at the same time, and the first clear, accurate answer usually wins. This skill drafts that answer fast without inventing anything.

## What you need

1. **The inquiry,** exactly as it arrived.
2. **The facts sheet.** If there isn't one, offer to build it with the facts-sheet skill, or ask for the three facts this reply needs: what they offer, how booking works, and the next step.
3. **Optional:** two or three past replies, so the draft sounds like them.

## Steps

1. List what the person actually asked. There are often two questions in one message.
2. Match each question to a line on the facts sheet.
3. When a fact isn't on the sheet, don't fill it in. Say it will be confirmed, and when.
4. Draft the reply.

## The reply

- Under 80 words. People read replies on their phones.
- Answer their question first, before any greeting paragraph or bio.
- One next step, with a date or a link. Not three options.
- Their voice: plain and warm, no hype.

If the inquiry came in on a social app, add a shorter DM or text version.

## The four checks

Show these under the draft, ticked or flagged:

- [ ] Answered their question first
- [ ] One next step, with a date or a link
- [ ] Nothing beyond the facts sheet (flag any price, date, promise or discount that isn't written down)
- [ ] Under 80 words

## The traps

- **Availability.** You can't see their calendar, so never confirm a date. Write "let me check and confirm by [time]".
- **Price.** If prices aren't public on the facts sheet, follow their rule (for example, send the price list after they share an email).
- **Exceptions and discounts** are always the musician's call.

## Auto-reply mode

When they want text for a form or CRM auto-reply, write a template with fields in square brackets ([first name], [date asked about]). It should answer the most common question, say a personal reply is coming, and give one next step.

The user sends every message. This skill only drafts.

*From the Blues People AI Agency course, chapter 4: Capture.*

---
name: follow-up-sequence
description: Write the follow-up messages a music business sends after an inquiry (the 72-hour date hold, the booked confirmation, the after-the-gig thank-you and review ask, and a 7-day nudge for leads who went quiet), with timing and the rule for when each one stops. Use whenever a musician asks what to send after an inquiry, how to follow up without being pushy, needs email or CRM automation copy, or has leads that stopped replying.
---

# Follow-up sequence

The ones who don't reply aren't gone, they're busy. Following up is a job, and you already have one, so the follow-up runs on a clock instead of on your memory.

## The sequences

| Sequence | Starts when | Messages |
| --- | --- | --- |
| A. Date hold | Someone asks about a date and gets your first reply | 24 hours, 48 hours, 72 hours (the hold ends) |
| B. Booked | The agreement is signed and the deposit is paid | Confirmation, what happens next, what you need from them |
| C. Reactivation | A lead has been quiet for 7 days | One friendly nudge with one easy next step |
| D. After the gig | The day after the performance | Thank you, a review or testimonial ask, the rebook line |

Write only the sequences the user asks for, and ask which ones if they don't say.

## What you need

- The facts sheet, or at least the facts these messages touch: how holds work, what booked means, the booking link.
- **Their definition of booked,** in one sentence. A common one is a signed agreement plus a deposit paid. Sequence B only starts on that.
- **Whether they really hold dates.** If they don't, Sequence A must not say "your hold ends today". False urgency breaks trust.

## Each message

- Under 60 words, warm, one clear next step.
- A short subject line when it's email.
- Fields in square brackets for the CRM: [first name], [event date], [booking link].
- Every sequence stops the moment the person replies.

## Output

A table the user can paste into a CRM or email tool:

| Sequence | When | Channel | Subject | Message | Stops when |
| --- | --- | --- | --- | --- | --- |

Then a few plain lines on setup: what starts each sequence and what stops it, in words they can follow in any tool.

## Checks

- Facts only, from the facts sheet.
- No fake deadlines or scarcity.
- The review ask comes after the gig, never before.
- Nothing sends from this skill. The user loads it into their own tool.

*From the Blues People AI Agency course, chapter 5: Follow-up, booking and getting paid.*

---
name: music-news-post
description: Turn a music-industry or AI-music news story into a short, accurate post for working musicians, with the fact, its number and its source, what it means for someone who plays for a living, and one either-or question that earns comments. Use whenever a musician shares a link or headline about streaming, royalties, AI music, labels, copyright, platforms or gear and wants to post about it on LinkedIn, Instagram, Threads or in a newsletter.
---

# Music news post

A good news post does three things: it states the fact cleanly, says what it means for someone who plays for a living, and asks a question people actually want to answer. It never guesses.

## Steps

1. **Read the source.** Open the link the user gave. If you can't open it, say so and work only from text they paste. Never write from a headline alone.
2. **Pull the facts:** who, what, when, and the one number that matters, with the outlet's name.
3. **Check it's fresh.** Note the date, and say so if it's more than a week old.
4. **Write it.**

## The post

1. **Hook:** one line that names the tension, not the headline.
2. **The fact:** what happened, with its number and the source named in the sentence.
3. **For working musicians:** two lines on what changes for someone who gigs, releases or teaches.
4. **The fork:** one either-or question with a time frame. "Does X happen before Y, or does Z?" beats "What do you think?"
5. **Call to action:** only if the user gives one, as one line at the end.

## Versions

- **LinkedIn:** up to about 1,300 characters in plain paragraphs.
- **Instagram:** a caption, plus five carousel slide lines (hook, the fact, why it matters, the fork, the call to action). The slides should carry the idea with the sound off.
- **Threads or X:** the hook, the fact and the fork in under 280 characters.

## Checks

- Every number appears in the source, with its unit.
- Predictions are framed as questions, never stated as facts.
- No invented quotes, and nobody named who isn't in the source.
- If the user voices it with an AI voice clone, the post says so.
- No hype words such as "game-changer" or "revolutionary".

*From the Blues People AI Agency course, chapter 6: The content engine.*

---
name: content-batch
description: Plan and draft a week of posts for a musician from things they already did (gigs, rehearsals, releases, lessons, the tools they use), in repeatable formats ("3 ways", "5 steps", an S-to-C tier list, a story post), laid out as a queue they approve once a week. Use when a musician says they never know what to post, wants a content calendar or a carousel, wants to turn gigs into content, or just says "help me post more", even if they don't mention a queue.
---

# Content batch

The weeks you gig hardest are the weeks you post least, which is exactly backwards. This skill turns the week you already had into a batch of posts you approve once, so posting stops depending on remembering.

## The rules this rests on

- **One queue.** Every post lives as one row in one place (Notion, a sheet or a doc). The row is the post.
- **One approval pass a week.** A status column reads Draft, Approved, Scheduled or Posted, and nothing ships until it says Approved.
- **One publisher per account.** One tool schedules each account. Two tools on one account is how posts vanish without anyone noticing.

## Step 1: inventory the week

Ask five quick questions:

1. Where did you play or record this week?
2. What did you practise, write or learn?
3. What went wrong, and what fixed it?
4. What tool or trick saved you time?
5. What's coming up that people can come to or buy?

## Step 2: pick formats

| Format | Shape | Good for |
| --- | --- | --- |
| 3 ways | Hook, three ways, call to action | Tips, fixes, gear |
| 5 steps | Hook, five steps in order, call to action | A process people can copy |
| S, A, B, C tier list | Hook, ranked items with one-line reasons | Tools, songs, venues, habits |
| Story | A moment, what it taught, a question | Gigs, mistakes, wins |

Pick three or four posts for the week and rotate formats so no two in a row match.

## Step 3: draft

- **Carousels:** slide 1 is the hook, then one idea per slide in under 25 words, and the last slide is the call to action. The slides should carry the idea with the sound off. Add a caption of three short paragraphs that ends in a question.
- **Video posts:** a caption plus a "record this" note saying what to film, 15 to 45 seconds. At least one post a week should have the musician on camera, because live footage of you does far more than a cloned voiceover.

## Output: the queue

| Date | Platform | Format | Hook | Status |
| --- | --- | --- | --- | --- |

Status starts at Draft. Put the full text of each post under the table, headed by its hook.

## Checks

- Real details only: venues, songs and numbers they gave you.
- One call to action per post.
- No client or venue named without permission.
- The user approves. This skill never posts or schedules.

*From the Blues People AI Agency course, chapter 6: The content engine.*

---
name: outreach-and-replies
description: Draft personal outreach for a musician (the monthly "what I've been building" note to past collaborators, follow-ups to bookers and venues, a show or release announcement) and sort the replies that come back into yes, stop, auto-reply and needs-you, with a ready answer for the clear yeses. Use whenever a musician wants to reconnect with contacts, announce something to their network, run an invite list, or is buried in replies to an email they sent.
---

# Outreach and replies

Outreach stays human. The tool builds the list, drafts the note and sorts the replies; the musician decides who hears from them and presses send. Outreach that gets started and never sent is the most common leak in a music business, so this skill keeps each send small enough to finish.

## Mode A: the outreach note

Ask:

1. Who is this for? Pick one segment: past collaborators, bookers, students, or fans on the list.
2. What's the one ask? Come to a show, listen to a release, reply "yes" for a free resource, or book a call.
3. What have you built, played or released lately? Two or three things.

Write:

- Under 150 words, from them, in their voice.
- One line at the top marked [personal line] for them to fill in per person. That line is what makes it a note and not a blast.
- One ask and one link.
- A last line that makes opting out easy: "Reply stop and I won't email you about this again."

Then advise: send it one at a time from their own address, or through their CRM to a tagged list. Start with five today, because a list of 70 that never goes out helps nobody. Never set an automation to message the people who book them.

## Mode B: the reply sorter

Sort each reply into one bucket:

- **YES:** a short yes and nothing else.
- **STOP:** unsubscribe, not interested, remove me.
- **AUTO:** out of office, bounce, auto-reply.
- **OTHER:** anything else, including a yes with a question attached.

Output:

| From | Bucket | Summary (OTHER only) | Next action |
| --- | --- | --- | --- |

- **YES** gets one fixed reply that the user wrote or approved. Draft it once and reuse it.
- **STOP:** remove them from every list today.
- **AUTO:** do nothing.
- **OTHER:** a one-line summary for the user to answer personally.

This skill drafts and sorts. It never sends.

*From the Blues People AI Agency course, chapter 7: Outreach and email.*

---
name: agent-workflow-writer
description: Turn a recurring task in a music business into a workflow an AI agent or automation can run without the musician in the room, covering the trigger, the action, where the output lands, how much freedom it gets, the one thing it must never do, and how you'll know it worked. Use whenever someone wants to automate a task, hand work to an AI agent or assistant, set up a scheduled task or a Zapier, Make or n8n flow, write instructions for a virtual assistant, or decide what to automate first.
---

# Agent workflow writer

A job an agent can run cold has three parts: what starts it, what it does, and where the result lands. Miss the third and you end up doing the agent's job of finding its own output. This skill writes the job so it runs without you.

## Step 1: the split

Before writing anything, sort the task:

| Column | Means | Examples |
| --- | --- | --- |
| Only me | Never handed off | Playing, recording, replies that take a stance, outreach to people who book you |
| Agent, with my veto | Runs, but I can stop it before it's public | Scheduled posts, drafted replies |
| Agent alone | Low risk, runs on its own | Filing notes, summaries, internal checks |

If it's "Only me", say so and stop there.

## Step 2: how much rope

Ask what happens if the agent gets it wrong. A small, private miss means it can run alone. Anything public, paid or permanent means it drafts and a person approves.

## Step 3: where it runs

From most to least reliable:

1. A scheduled cloud task that runs whether the laptop is open or not.
2. An always-on hosted service.
3. A task tied to a desktop app. It only runs while the app is open, so it's fine for chores and bad for deadlines.

The worst setup is two schedulers on one queue, each assuming the other did the work.

## Step 4: write the spec

```
Job: [name]
Trigger: [an event or a time, never "keep an eye on it"]
Inputs: [what it reads, and where from]
Action: [the steps, in order]
Output lands in: [a place the musician already reads]
Rope: [runs alone | drafts for approval]
Never: [the one thing it must not do]
Done when: [how you'll know it worked]
Check: [who looks, and how often]
```

## Checks

- The trigger is an event or a time, not a vibe.
- The output lands somewhere they already read. If they have to go looking for it, the job isn't finished.
- Numbers in the output come from a file or a live system, never typed from memory.
- Passwords, keys and card numbers never go into the instructions.
- Each channel has one publisher.
- Run it by hand once before scheduling it.

*From the Blues People AI Agency course, chapters 0 and 8.*

---
name: two-ai-cross-check
description: Check an important answer by giving the same question and the same context to two different AI models, then compare where they agree, where they differ, and how to settle it, and log the result. Use whenever a musician is about to trust an AI on numbers, a contract, a royalty statement, a chord chart, theory, gear specs, or any fact they'll publish or pay for, or asks "is this right?" about something an AI told them.
---

# Two-AI cross-check

One AI gives you an answer. Two AIs with the same context give you a check: agreement means you're probably in a good spot, and disagreement tells you exactly where to dig.

## Steps

1. **Package the context.** The same notes, files or screenshot go to both models. Different context makes the comparison meaningless.
2. **Write the question** in one or two sentences.
3. **Prompt both models with this:**

```
Here is my context: [notes, file or screenshot]. Here is the question: [question]. Answer it, then list the two claims in your answer you're least sure of and how I could check each one.
```

4. **Answer it yourself** as one of the two models, then ask the user to paste the other model's answer.
5. **Compare.**

## Output

| Claim | Model A | Model B | Agree? | How to check | Verdict |
| --- | --- | --- | --- | --- | --- |

Agree? is Agree, Partly or Disagree. Fill Verdict only once something outside both models settles it: the source document, a calculation, the score, the contract text, or a test.

Then one log line they can keep in a running table:

| Date | Question | Agree? | Who was right | How I checked |
| --- | --- | --- | --- | --- |

## Checks

- Agreement isn't proof, because two models can share the same mistake. Anything that will be published, signed or paid gets checked against the source.
- Say plainly when neither model can know the answer (private data, live prices, someone's calendar).
- Keep it about claims, not about which model is smarter.

*From the Blues People AI Agency course, chapters 8 and 10.*

---
name: stack-audit
description: Audit every tool and subscription in a music business (website, domain, email, CRM, scheduler, DAW and plugins, AI tools, storage, payments) with one test, what breaks if you pull it out tomorrow, and rank each one S, A, B or C with its cost, who owns the login, and what to cut or automate next. Use whenever a musician asks which tools they need, wants to cut subscriptions, is setting up their stack, or is rebuilding after a lost laptop, a hacked account or a dead login.
---

# Stack audit

Rank every tool by what breaks if you pull it out tomorrow. It's the fastest way to see what you actually depend on, what you're paying for out of habit, and which logins aren't even in your name.

## Step 1: the inventory

For each tool, ask:

- What job does it do?
- What does it cost a month? Unknown is fine; mark it.
- Whose name is the login in?
- Where does its data live, and could you get it back?
- What depends on it?

## Step 2: rank

| Tier | Means |
| --- | --- |
| S | The business stops without it |
| A | Real damage within a week |
| B | Annoying, but you'd manage |
| C | You wouldn't notice |

## Output

| Tool | Job | Cost a month | Login owner | Tier | Reason (one line) |
| --- | --- | --- | --- | --- | --- |

Then exactly three recommendations:

1. **Cut one:** a C-tier tool you're paying for.
2. **Automate one:** a manual step the stack could already do.
3. **Move one login into your own name,** starting with the domain and email.

## Red flags to call out

- A login in a former bandmate's, manager's or developer's name.
- Data that lives only inside one tool, with no copy anywhere you control.
- Two tools publishing to the same account.
- A tool everything depends on, with no backup or export.

## Checks

- No invented costs. Unknown stays "unknown".
- Every recommendation names the tool.

*From the Blues People AI Agency course, chapters 1 and 12.*

---
name: money-guardrails
description: Set the rules before any AI tool, agent or automation touches money in a music business (deposits, invoices, payment links, payouts, contracts with payment terms, billing code or bank data), producing a guardrail checklist, written definitions of "booked" and "paid", and the exact steps that need a human yes. Use whenever a musician wants to automate invoicing, deposits or contracts, connect payments to a CRM or website, or let an AI help with billing, even if they only ask "can AI do my invoices?"
---

# Money guardrails

The agent does the legwork. Money, secrets and deploys wait for a person. The bandleader holds the book and the pay envelope.

## Step 1: map where money moves

Ask where money comes in and goes out: deposits, final payments, payouts to players, refunds, subscriptions. For each flow, mark what an AI may do:

| Money flow | AI may see | AI may draft | Never touches |
| --- | --- | --- | --- |

## Step 2: write the definitions

- **Booked:** for example, a signed agreement plus a deposit paid. Their words, in one sentence.
- **Paid:** money received and cleared, not "they said it's sent".

## Step 3: the guardrails

1. **Turn the money off first.** When an AI or a developer works on anything connected to payments, switch off the live connections (payment webhooks, automations that charge) before the work starts. Turning them back on is a separate step that waits for a person.
2. **Test with no real money.** Use the payment tool's test mode until a person has checked the whole path end to end.
3. **Changes get reviewed before they go live.** Code or automation changes are proposed, a person approves them, then they go live.
4. **Keys stay with the human.** Live payment keys, bank logins and card numbers never go into a chat, a prompt or a shared doc. Say where a key lives; never paste its value.
5. **Check what ships.** Every download, template or file you give out gets checked for keys or private data first.

## Output

- The money map table.
- The two definitions.
- A checklist with a box for each guardrail, plus who says yes for every step that goes live.

## Red flags

Stop and tell the user if anything asks for a password, a full card number, a bank login or a live secret key in chat.

*From the Blues People AI Agency course, chapter 9: Money and guardrails.*

---
name: sunday-review
description: Run a musician's weekly Sunday review, gathering the week's notes, logs, bookings, inquiries and posts into one page that splits "This week, on you" (decisions with due dates) from "Agent queue" (what tools did or will do), plus three numbers (inquiries, replies sent within an hour, and bookings). Use whenever someone asks for a weekly review, wants to plan the week, feels behind or scattered, or wants one page instead of checking every app.
---

# Sunday review

Your tools write logs all week. You read one page on Sunday. This skill writes that page.

## Inputs

Ask for whatever they have: the calendar, inbox or CRM notes on inquiries, the content queue, task lists, automation logs and last week's review. Work with what you get and say what was missing.

## Steps

1. Pull every open item into one list.
2. Split it: does it need the musician's decision or action, or not?
3. Count the three numbers from the source (calendar, CRM, inbox), never from memory.
4. Look for what didn't happen: posts that didn't go out, follow-ups that didn't fire, an automation with no recent run.

## Output

```
# Week of [date]

## This week, on you
- [decision or action] (due [date])

## Agent queue
- [what a tool did or will do] (nothing for you unless marked NEEDS YOU)

## Three numbers
- Inquiries: [n]
- Replies sent within an hour: [n of n]
- Bookings (signed and paid): [n]

## What didn't happen
- [item, and how many days overdue]

## One suggestion
- [the single change that would help most next week]
```

## Checks

- "On you" holds only what truly needs them. If it's long, the split is wrong.
- Every number says where it came from. Unknown stays "not counted".
- One suggestion, not five.

*From the Blues People AI Agency course, chapters 2, 12 and 13.*

---
name: ai-part-check
description: Check a part an AI music tool produced (chords, a bass line, a melody, a vocal harmony, a lead sheet or a generated track) against a written spec of key, chords, form, tempo and range, and report plainly what it got right, what it got wrong and how far above chance it is. Also helps write that spec before generating. Use whenever a musician uses Suno, Udio, Magenta, ElevenLabs, a DAW's AI assistant or any model to write or perform music and wants to know if it's actually right.
---

# AI part check

Treat a music model like a new player on the gig: hand it a chart, then check what it played against the chart. Without a chart you can only judge the vibe, and models sound confident when they're wrong.

## Step 1: the spec, before generating

Help the user write it:

```
Key: [e.g. Bb major]
Tempo and feel: [e.g. 92 BPM, swing]
Form: [e.g. AABA, 32 bars]
Chords: [one per bar or per beat, e.g. | Bbmaj7 | G7 | Cm7 | F7 |]
Part: [bass | melody | harmony | comping]
Range: [lowest and highest note]
Must: [e.g. land on the 3rd of each chord on beat 1]
```

## Step 2: get the notes

The check needs notes, not just audio. MIDI or notation from the tool is best, the user's own transcription is next. If there's only audio, say plainly what can't be verified from here and suggest an audio-to-MIDI step.

## Step 3: check

For each bar, compare the notes with that bar's chord in the spec:

- **Chord tones:** the share of notes that belong to the chord.
- **In key:** the share of notes inside the key.
- **Form and length:** the right number of bars and sections.
- **Range:** any note outside it.

**Compare with chance.** Random notes land on a 4-note chord about 33% of the time (4 of 12 pitch classes) and inside a 7-note scale about 58% of the time. A score near those floors means the model isn't really following the chart, however good it sounds. Show the floor beside every score.

## Output

| Bar | Chord | Notes played | Chord tones | Issues |
| --- | --- | --- | --- | --- |

Then a summary: the chord-tone share against the 33% floor, the in-key share against the 58% floor, any form or range problems, and the two bars to fix first.

## Notes

- Label any AI vocal or voice clone as AI wherever it's released.
- If the user's AI setup can call a harmony tool (an MCP server such as THIRI, with a free first key at build.thiri.ai/keys), the agent can ask it for chords and voicings instead of guessing. Mention it only as an option.
- Pitch accuracy and harmonic correctness are different checks. A part can be perfectly in tune and still sit on the wrong chord.

*From the Blues People AI Agency course, chapter 10: AI in the music itself.*