run-ads
Skill1 starsChanged 3 months agoMITView on GitHub
Run a paid advertising campaign end-to-end — economics check, account structure, creative, launch, pacing, kill-or-scale, close-out. Run when strategy calls for paid spend and something has already proven it converts.
In this repository22 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- build-funnelSkill
- close-dealSkill
- connect-toolsSkill
- create-contentSkill
- define-positioningSkill
- design-offerSkill
- experimentSkill
- grow-referralsSkill
- grow-seoSkill
- health-checkSkill
- launch-campaignSkill
- launch-productSkill
- manage-pipelineSkill
- measureSkill
- plan-strategySkill
- research-marketSkill
- review-monthlySkill
- run-outboundSkill
- run-partnershipsSkill
- setup-teamSkill
- tune-revopsSkill
The file
Files
/run-ads — Spend Carefully, Learn Fast
Inputs: knowledge/strategy.md (paid must be a chosen channel), knowledge/budget.md (ceiling + approval ref), knowledge/offer.md, knowledge/audience-profiles.md, knowledge/research/social-proof.md, knowledge/tool-registry.md.
Agents: paid-ads-specialist leads; copywriter writes ad copy; funnel-builder confirms the destination; tools-operator configures the platform; growth-analyst verifies tracking and reads results.
Method: knowledge/methods/paid-advertising.md.
Steps
- Prove it deserves spend. Has something converted organically? Run the CPA-ceiling arithmetic (method §1): customer value × acceptable share = ceiling; ÷ destination conversion rate = maximum affordable click. If the platform's going rate exceeds it, stop here and report why — this is a legitimate and valuable outcome of this skill.
- Confirm budget and approval. Headroom in
knowledge/budget.md, a covering money-gate approval ref, and a test budget small enough to lose while learning (typically $10–20/day for 7–10 days) that still buys a readable result (≥10 conversions, ideally 30+). - Pick the platform by intent. Retargeting first if traffic supports an audience; search if real intent exists; cold social last. Name the platform not chosen and why.
- Ready the plumbing. tools-operator confirms account access, pixel/conversion tracking installed, and the conversion event set to the real business outcome — verified by completing the action personally. Untested tracking is the most expensive bug in this channel.
- Brief and build the creative.
templates/ad-brief.md: audience, one promise, one reason to believe (real proof fromknowledge/research/social-proof.md), one ask. Three genuinely different concepts, not three headline variations. Visuals via Canva againstknowledge/brand-assets.md. Claims Review Checklist plus the platform's own ad policy before anything uploads. - Write the kill rules, then launch. Kill/refresh/scale conditions into the brief before the campaign goes live (method §5). Structure per the naming convention so the campaign reconciles to its file. Launch; check within 24 hours that it's delivering and the conversion event is firing.
- Pace, don't fiddle. Days 2–3: watch for catastrophe only — editing during the learning phase resets it and wastes the spend that bought the learning. After the learning phase: cost per result vs. ceiling. Weekly: reconcile actual spend to
knowledge/budget.md(the bank is authoritative, not the dashboard). - Kill or scale on the rule. Execute the pre-committed condition without renegotiating it. Scaling means ~20% increments with time to stabilize, never doubling.
- Close out. Final numbers vs. the ceiling, what the creative test taught, honest attribution (with what couldn't be attributed said plainly), and one dated learning line to
knowledge/methods/paid-advertising.md.
Before you start
- Required inputs:
knowledge/strategy.mdapproved with paid as a chosen channel;knowledge/budget.mdwith an approved allocation; a live, instrumented destination (else/build-funnelfirst). - Scan
knowledge/decision-log.mdfor tagsbudget,channel— a previously killed platform or angle isn't retried without naming what changed. - Gate check: money gate is mandatory and verified at step 2, not assumed. Ad creative is public — the publishing gate and the Claims Review Checklist both apply.
If things go wrong
- The economics don't clear at step 1 → report that plainly and recommend where the money goes instead (usually outbound for B2B, organic content or retargeting-only for B2C). A campaign that shouldn't run is a successful outcome for this skill.
- Conversion tracking can't be verified → do not launch. Spending into unmeasurable outcomes is spending blind; park in
knowledge/pending.mduntil it's fixed. - Ads rejected on policy → read the specific policy line cited, fix that specific thing, resubmit once. Repeated rejections risk the account — escalate rather than retrying variations.
- Spend outpaces the approved cap → pause immediately and escalate. Overspend is a gate violation, not a rounding error.
- Platform account requires identity/payment verification → guided manual steps for the operator; park until done.
Outputs
A launched, paced, and eventually closed-out paid campaign at knowledge/campaigns/YYYY-MM-<slug>.md — or a documented recommendation not to spend. Close with the report per templates/skill-report.md.
Writeback
Campaign file; spend and return to knowledge/budget.md; results to knowledge/results-history.md; approval refs to knowledge/approvals-log.md; kill/scale decisions (tags: budget, channel) to knowledge/decision-log.md; a learning line to knowledge/methods/paid-advertising.md.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

