Ad Spend Auditor

What it does

WASTE
| Item                         | Spend | Results        | Why flagged                     |
| ---------------------------- | ----- | -------------- | ------------------------------- |
| Prospecting / Lookalike 1%   | $212  | 0 conversions  | Over $50 threshold, no results  |
| Prospecting / Interest Stack | $64   | 0 conversions  | Over $50 threshold, no results  |

FATIGUE
| Creative            | Days live | CTR trend      | Frequency trend |
| ------------------- | --------- | -------------- | --------------- |
| "Summer Sale" video | 42        | 2.1% -> 0.8%   | 1.4 -> 3.9      |

REFRESH ANGLES
- Prospecting / Lookalike 1% ($212): pause it; test a 2% lookalike.
- Prospecting / Interest Stack ($64): pause it; test one narrower interest.
- "Summer Sale" video: retire it; test a new hook on the same offer.

SUMMARY
The biggest burn is Prospecting / Lookalike 1%: $212 of the $276 flagged
spend (77%) produced no conversions.

What this one touches

Reads only the ad spend export you hand it; it never touches your ad account. Writes a report and nothing else. Never changes bids, budgets, audiences, or campaign status.

Install

Copy this block and paste it into your Muse chat. Muse does the install and confirms when it's done.

Full source

This is the exact text Muse installs. Read it. That's the whole trust model.

---
name: ad-spend-auditor
description: Audit an ad spend export for wasted spend and creative fatigue, then suggest refresh angles. Trigger phrases: "audit my ad spend", "review my ad spend", "find wasted ad spend".
---
# Purpose

Find the specific places where ad money is burning: campaigns with
rising cost per result and no conversions, ad sets spending on
narrow segments that never convert, and creative that has run so
long its frequency is climbing while its click-through rate falls.
Report waste first and refresh ideas second, all from one export.

# Workflow

1. Read the ad spend export: a CSV or pasted table with at least
   campaign, ad set, ad name, spend, impressions, clicks,
   conversions, and the date range. Ask which platform and range if
   either is unclear.
2. Compute per campaign and per ad set: spend, impressions, CTR,
   cost per click, cost per result, and conversion count.
3. Flag spend without results: anything over the user's waste
   threshold (ask for one; default $50) with zero conversions.
4. Flag fatigue: ads where frequency is rising and CTR is falling
   across the range, or the same creative live more than 30 days
   with falling performance.
5. Flag overlap waste: ad sets on the same platform targeting
   near-identical audiences with near-identical creative.
6. For each flagged item, draft one concrete refresh angle: a new
   hook, a new audience slice, or a pause recommendation, grounded
   in the numbers.
7. Write the report in the Output Contract shape. Never adjust a
   budget or pause anything; analysis only.

# Output Contract

- Waste table: item, spend, results, and why it is flagged.
- Fatigue table: creative, days live, CTR trend, frequency trend.
- Refresh angles: one short line per flagged item, most spend
  first.
- One-paragraph summary naming the single biggest burn.

# Operating Rules

- Never change bids, budgets, audiences, or campaign status;
  read-only analysis.
- Never invent numbers; every figure comes from the export.
- Ask for missing columns before guessing.
- Keep the report short enough to act on in one sitting.

More in Marketing

• Review Responder

  Turns customer reviews, good and bad, into on-brand public replies that defuse complaints and make promoters feel seen.

  skillinstructionsworkflowreply-templatestone-calibration

  by Muse Skills editors · Written for this directory

  ViewCopy install promptInstall the "Review Responder" skill. Its full source is below. Create it at ~/workspace/skills/review-responder/SKILL.md following skill-creator conventions (name and description frontmatter; Purpose, Workflow, Output Contract, Operating Rules sections). Then confirm it is installed and tell me the trigger phrases. --- SOURCE --- --- name: review-responder description: Drafts on-brand public replies to customer reviews: defuses bad reviews, thanks good ones, and moves real problems offline. Trigger phrases: "draft a reply to this review", "respond to this bad review", "write review responses", "review reply". --- # Purpose Do one narrow job: turn raw customer reviews into public replies worth publishing. The skill takes the review text, the star rating, the business name, and your tone, then drafts a reply that sounds like a real human who read the review. For negative reviews it acknowledges, takes the fix offline, and never argues in public. For positive reviews it thanks specifically, names something the reviewer mentioned, and encourages a return visit. You always review and post; the skill only drafts. # Workflow 1. Collect the inputs: the review text, star rating, platform (Google, Yelp, Amazon, TripAdvisor, app store), business name, and your desired tone (warm and casual, professional, playful, formal). 2. Classify the review: praise, complaint, mixed, or spam/suspicious. Spam and policy-violating reviews get a brief factual reply plus a flag-and-report recommendation, not a full engagement. 3. For complaints: open by naming the specific issue (never a generic "sorry for your experience"), take responsibility for the part that is yours, give a private contact path (name, phone or email placeholder), and invite them to reach out. Never argue the facts in public, never admit legal fault, never promise refunds or compensation in the reply. 4. For praise: thank them by naming the specific thing they liked, add one line of personality, and invite them back. Skip the hard sell; a genuine reply earns more than a coupon. 5. For mixed reviews: acknowledge the positive, address the negative with the complaint playbook, and keep the reply shorter than the review. 6. Produce two versions: the recommended reply and a shorter alternate. Both stay under 100 words unless the platform norms differ. Ask for the business details you need before drafting. # Output Contract - The review classification (praise, complaint, mixed, or spam) - A recommended reply under 100 words, ready to copy - A shorter alternate version - A one-line note on what to fill in (contact name, phone, order details) before posting - For spam: a brief reply plus a flag-and-report recommendation # Operating Rules - Never post anything to a review site; you draft, the owner posts. - Never include private customer details in a public reply. - Never promise refunds, discounts, or compensation in a reply unless the owner told you to. - Never argue with the reviewer in public, even when they are wrong. Take it offline. - Never invent facts about what happened. If the complaint is vague, ask for context or keep the reply general and forward it to a private channel. - Match the tone the owner gave you; keep the voice human, not corporate filler.

• Ad Hook Researcher

  Finds the ad hooks already winning in your niche and turns them into a creative brief.

  skillinstructionsworkflowbrief-template

  by Muse Skills editors · Written for this directory

  ViewCopy install promptInstall the "Ad Hook Researcher" skill. Its full source is below. Create it at ~/workspace/skills/ad-hook-researcher/SKILL.md following skill-creator conventions (name and description frontmatter; Purpose, Workflow, Output Contract, Operating Rules sections). Then confirm it is installed and tell me the trigger phrases. --- SOURCE --- --- name: ad-hook-researcher description: Collect winning ad hooks in a niche from a product URL into a creative brief. Trigger phrases: "research ad hooks", "find winning hooks", "build an ad hook brief". --- # Purpose Find the ad hooks already winning in your niche and hand them to you as a creative brief you can work from. You give a product URL (or a niche description), and Ad Hook Researcher returns a hook bank, the patterns behind the hooks, the angles nobody is using, and three starter hook drafts written for your product. # Workflow 1. Read the product URL or niche description. If you gave both, use the page to ground the niche read; if you gave one, start there. 2. Read the product's own landing page: headline, subhead, hero claims, pricing framing, and the promises on the page. This is the baseline the brief adapts hooks for. 3. Search the niche for active ads: public ad libraries, competitor landing pages, and ad creative roundups. Note which ads or hooks keep showing up across sources; persistence is an observed signal, not proof of performance. 4. Extract each hook: the first five to ten words plus its angle. Tag the angle: transformation, curiosity gap, direct callout, proof, offer, or contrast. 5. Cluster the hooks by angle and note which clusters dominate the niche and which angles are missing entirely. 6. Keep only hooks from ads that were actually found. Never invent ad text. Quote short hooks verbatim; summarize longer ones. 7. Write three starter hooks for the product, clearly marked as drafts, each using a different winning angle from the niche. 8. Write the creative brief in the Output Contract shape and send it. # Output Contract The creative brief, in this order: - Niche snapshot: what your page leads with today, in three lines - Hook bank: 10 to 15 hooks, each with an angle tag and the source type it was found in - Pattern notes: three to five recurring patterns across the niche - Gaps: angles the niche is not using, with one line on why each could fit - Starter hooks: three drafts for your product, marked as drafts, each using a different winning angle - Sources checked: a short list of what was actually read # Operating Rules - Read public pages and public ad libraries only. Never log in, and never work around a paywall or a bot check. - Never present persistence as proof of performance. Say what was observed. - Never invent ad copy. Every hook in the bank comes from a real page or ad that was read during the run. - Never buy media, post an ad, or contact a competitor. Research only. - If the niche is too thin to fill the hook bank, deliver what was found and say what is missing rather than padding it. - Run once per request; do not re-run on a schedule unless asked.

• Lead Scorer

  Scores a list of public leads against your ideal customer profile and ranks them.

  skillinstructionsworkflowscoring-rubric

  by Muse Skills editors · Written for this directory

  ViewCopy install promptInstall the "Lead Scorer" skill. Its full source is below. Create it at ~/workspace/skills/lead-scorer/SKILL.md following skill-creator conventions (name and description frontmatter; Purpose, Workflow, Output Contract, Operating Rules sections). Then confirm it is installed and tell me the trigger phrases. --- SOURCE --- --- name: lead-scorer description: Score public leads against your ideal customer profile and rank them. Trigger phrases: "score these leads", "rank my leads", "which leads first". --- # Purpose Stop working your lead list in alphabetical order. Lead Scorer takes the list you hand it, scores each lead against the ideal customer profile you define, and returns a ranked list with the evidence behind every score. You see exactly why a lead sits at the top or the bottom, so your outreach starts with the people most likely to be a fit. # Workflow 1. The user provides the ideal customer profile: industry, company size, role titles, signals that suggest fit (hiring, funding, tech stack), and deal breakers. 2. The user provides the lead list: names plus whatever public data they have per lead (company, title, location, public posts, listing details). 3. Score each lead 0 to 100: fit signals add points, missing basics subtract, deal breakers zero the score. Record the evidence for every point. 4. Rank the list high to low. Group obvious tiers: reach out now, nurture later, skip. 5. Deliver the ranked list with one evidence line per lead, plus the rubric used, so the user can reweight it. # Output Contract Each run delivers: - The ranked lead list with 0-100 scores - One evidence line per lead showing why it scored that way - Three tiers: reach out now, nurture later, skip - The rubric, written out, so scoring stays repeatable A score is a ranking aid, not a verdict. The user reviews the evidence before acting on it. # Operating Rules - Score only public data and what the user provides. Never scrape private accounts, and never use bought or breached data. - A deal breaker zeroes the score but keeps the lead visible, flagged, with the reason shown. - Never inflate a score to be encouraging. A weak list gets weak scores, honestly. - The rubric is fixed for the whole list. Never reweight mid-list to favor one lead. - Draft no outreach and contact nobody. Scoring ends where outreach begins, and a human writes and sends every message. - Keep the user's lead list private to their workspace.

• Prospect Reply Drafter

  Reads a prospect's reply in the context of the thread and drafts the right next step.

  skillinstructionsworkflowtemplates

  by Muse Skills editors · Written for this directory

  ViewCopy install promptInstall the "Prospect Reply Drafter" skill. Its full source is below. Create it at ~/workspace/skills/prospect-reply-drafter/SKILL.md following skill-creator conventions (name and description frontmatter; Purpose, Workflow, Output Contract, Operating Rules sections). Then confirm it is installed and tell me the trigger phrases. --- SOURCE --- --- name: prospect-reply-drafter description: Read a prospect's reply in the context of the thread and draft the right next step. Trigger phrases: "draft a reply to this prospect", "how should I respond", "write a follow-up for". --- # Purpose Read a prospect's latest reply together with the full thread history, figure out what the prospect is actually signaling, and draft the single best next move. It classifies the reply, an objection, a timing deferral, a referral, a polite no, or genuine interest, and writes a response matched to that signal instead of a generic follow-up. # Workflow 1. Read the full thread: your outreach, their reply, and any earlier exchanges. Ask for the offer or call to action if it is not clear from the thread. 2. Classify the reply: interested, objection, timing deferral, referral, polite no, or ambiguous. 3. Identify the signal underneath the words: what they care about, what is blocking them, what would change their mind. 4. Draft one reply: short, specific, and matched to the classification. Handle objections by addressing the block, not by arguing. 5. Give the human a send-or-edit choice: the draft plus two alternate lines for the risky spots. # Output Contract - Classification: one label plus the evidence line from the reply. - Signal read: what the prospect cares about and what blocks them, in two bullets. - Draft reply: subject or opener plus body, ready to paste, under 120 words. - Alternates: two rewrite options for the key line. # Operating Rules - Never send anything; the human reviews and sends every reply. - Never invent facts about the product, pricing, or the prospect. - Never guilt-trip, never pressure, never fake a personal connection. - If the reply is a no, draft a graceful close, not a rescue attempt.