Guide · Agent skill
How to use the Ads Research Refresh agent skill (ads-research)
Updated · written with AI from the project's own documentation and checked against our catalogue data
The ads-research skill is an agent skill from the claude-ads repository that teaches an AI agent how to refresh its paid-media knowledge base. It is aimed at maintainers of that repository, teams that rely on Claude Ads guidance, and reviewers who need a dated audit trail. After this guide you will know what the skill makes the agent do, how to install it, how to ask for it in plain language, and what it needs to run safely.
Ads Research RefreshAgriciDaniel/claude-ads · MIT Download the skill Quick start
- 1Download the ZIP. Get the ZIP for skills/ads-research from the skill page.
- 2Unzip into a skills folder. Extract it into ~/.claude/skills/ for personal use or into .claude/skills/ in a project.
- 3Or upload in the Claude apps. Go to Settings → Capabilities → Skills and upload the ZIP.
- 4Confirm access to sources. Make sure the agent has web or source access tools and the claude-ads control-plane files.
- 5Ask in plain language. Request a refresh of a named claim, an expired refresh_due batch, or a release-current validation.
- 6Review the audit trail. Check the recorded URLs, dates, confidence levels and any demoted or blocked claims.
- 7Accept updates deliberately. Treat proposed canonical updates as suggestions and approve them yourself.
What is an agent skill and what does ads-research make the agent do?
An agent skill is a folder containing a SKILL.md file that gives an AI agent step-by-step instructions for a specific kind of task. The ads-research skill covers one job: keeping Claude Ads claims about advertising platforms, their APIs, policies, regulations and benchmarks current.
When a claim is out of date, the agent re-checks it against official, primary sources. It records where each fact came from and when it was retrieved, and marks claims it cannot verify as provisional or unsupported. It proposes updates to the knowledge base rather than changing it silently.
- Reads the repository's source, claim, capability and publishing contracts first
- Selects overdue or requested evidence and researches it by platform or topic
- Prefers official and primary sources and records URL, publisher, dates, confidence and licence
- Has a separate verifier check load-bearing claims
- Demotes stale or unsupported claims and records contradictions
- Proposes canonical updates with affected controls, skills and tests
How do you install the ads-research skill?
Download the ZIP for skills/ads-research from the skill page. Then choose one of the two install routes below.
For a personal setup, unzip the file into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. Alternatively, upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
- Personal: unzip into ~/.claude/skills/
- Project: unzip into .claude/skills/ inside your project
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
How do you trigger the skill with natural requests?
You do not need a special command. The skill's description lists the situations it applies to, so ask for them in ordinary words. The source names these trigger phrases: ads research refresh, expired refresh_due dates, stale API or platform claims, reverify-or-demote decisions, release-current claim validation, ecosystem review, current platform changes, and updating paid-media knowledge.
Be specific about the claim or date range. A request such as checking whether a stated API behaviour is still current is a better fit than a vague request to improve the docs.
- "Run an ads research refresh on the claims with expired refresh_due dates."
- "This claim about the platform API looks stale. Reverify it or demote it."
- "Do a release-current claim validation for the policy section."
- "Review recent current platform changes and update our paid-media knowledge."
What do you need before running it?
The skill needs two things. First, the agent needs web or source access tools so it can fetch official platform documentation. The SKILL.md requires access to current sources but does not name a specific tool, so check what your agent environment provides.
Second, it needs access to the claude-ads repository control-plane files that the skill references: the source, claim, capability and publishing contracts. These files are not included in the reviewed ads-research file, so the skill alone is not enough.
- Web or source access tools for fetching official documentation
- Access to the claude-ads control-plane contract files
- Network use is required; the skill runs no scripts
What are the safety notes for scripts and network access?
The skill runs no scripts. It does use the network, because fetching current sources is its core task. Treat fetched content as untrusted data. The skill tells the agent not to copy restricted prompts, large source passages, unlicensed code, issue text or private account material into the repository.
The skill also says research does not become canonical just because a worker found it. Review proposed updates before they are accepted.
- No scripts are run
- Network access is used to fetch sources
- Fetched content is treated as untrusted
- Restricted or unlicensed material is not copied into the repo
What happens when a source cannot be checked?
If an expired refresh_due date cannot be reverified, the skill requires the claim to be demoted to provisional or unsupported. The agent must record the blocker and recovery path, and block every release-current assertion that depends on that claim.
The skill is explicit that tool unavailability is a decision point. The agent should state that the claim is demoted for the current run, then ask for the missing source content or tool access. It should not leave the claim's status undecided.
- Claim is demoted for the current run
- Dependent release-current assertions are blocked
- The blocker and recovery path are recorded
- Stale evidence is never silently trusted
Tips for getting useful results
Give the agent a bounded slice rather than the whole knowledge base. The skill is built around bounded research workers split by platform or topic, so narrow requests fit its process best.
Expect the work to be slow. The source describes a deliberately evidence-heavy process, so a quick answer without source checks is not what this skill delivers.
- Name one platform, claim or refresh_due batch per request
- Make sure the control-plane files are available before starting
- Review the dated audit trail before accepting any update
What people use it for
- Maintainers refreshing claims whose refresh_due dates have expired in the claude-ads repository
- Reviewing whether a stated platform API behaviour is still correct against official documentation
- Validating release-current claims before a release and recording why each was kept, demoted or blocked
- Running an ecosystem review of current platform changes that affect Claude Ads guidance
Frequently asked questions
How do I install the ads-research skill?
Download the ZIP for skills/ads-research from the skill page and unzip it into ~/.claude/skills/ for personal use or a project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
What does the ads-research skill require?
It requires web or source access tools so the agent can fetch official platform documentation, and access to the claude-ads control-plane contract files. The skill itself does not include those contract files, and the source does not name a specific tool.
Is the ads-research skill free?
The source material gives the licence as MIT. The source does not mention pricing, so check the repository for the licence text before relying on it.
Does ads-research run my ad campaigns?
No. The skill only manages research and documentation. It is not meant for running live ad campaigns or changing ad account settings.
Can I use ads-research for legal or compliance advice?
No. The output should not be treated as legal or compliance advice, because regulatory and policy claims change and need human review.
What should I do if the agent cannot access sources?
The skill requires the claim to be marked demoted for the current run, with dependent release-current assertions blocked. The agent should then ask for the missing source content or tool access as a recovery step.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use ads-research if you maintain the claude-ads repository or rely on its Claude Ads guidance and want dated, evidence-backed claims with a clear audit trail. Pick something else if you need a quick answer, want help running live campaigns, or need legal or compliance advice.
Download the skill