How to use the Validate Claude Ads Releases agent skill

What an agent skill is and what ads-validate makes the agent do

An agent skill is a folder with a SKILL.md file that gives an AI agent specific instructions for a task. When a request matches the skill's description, the agent reads those instructions and follows them.

ads-validate turns the agent into a quality checker for the Claude Ads repository. It picks the narrowest check that fits the request: a contract or bundle, scores, a run, a capability, the whole repository, or a release. It returns machine-readable pass or fail results, names the highest-priority blocker, and suggests recovery steps.

• Contract or bundle: uses the deterministic core validator.
• Scores: checks controls, findings, category weights and coverage.
• Run: checks manifest completeness, source lineage, privacy, worker status and render artifacts.
• Capability: requires implementation, fixtures, tests, sources and truthful mode.
• Repository: runs deterministic, routing, security, installer, rendering, evidence, packaging and freshness gates.
• Release: dispatches a fresh-context release verifier.

Install ads-validate in Claude

Option 1, personal install: download the ZIP from the skill page on this site and unzip it into ~/.claude/skills/. The skill folder should then sit in that directory so the agent can find its SKILL.md.

Option 2, project install: unzip the ZIP into the .claude/skills/ folder inside your project, so the skill applies only to that project.

Option 3, Claude apps: download the ZIP, then upload it in the Claude apps under Settings → Capabilities → Skills.

• Use ~/.claude/skills/ for a skill you want in every project.
• Use a project's .claude/skills/ folder to keep it scoped to one repository.
• Start a new session after installing so the agent loads the skill.

Trigger ads-validate with natural requests

You do not need a special command. The skill's description lists the phrases it responds to, so ask in plain language that names what you want checked. The more specific the target, the narrower and faster the check.

The skill's description includes these trigger phrases: ads validate, ads status, ads next, maturity checks, ownership-manifest uninstall, preserving unrelated ads-* skills, checksum verification, preflight, QA and release audits.

• Validate the contract for the latest run bundle and tell me the top blocker.
• Check whether the install and uninstall behaviour preserves unrelated ads-* skills.
• Run a preflight on the repository before I tag a release.
• Verify the checksum of the release archive against the trusted release channel.
• Give me ads status and ads next for this capability.

Understand the results and safety rules

Results are pass or fail with the evidence behind them. The skill says not to raise a maturity level because documentation looks polished or because a previous release passed. Stale evidence and skipped remote CI lower readiness.

For install and uninstall, the skill only allows deleting exact paths listed in the matching ownership manifest. It validates the whole manifest and the configured root boundaries before any deletion. If the manifest is missing, invalid, mismatched or unsafe, the agent must stop and ask for manual review.

The skill rejects pipe-to-shell install instructions. Archives need a SHA-256 checksum verified against a trusted release channel before they are run.

• It never finds uninstall targets with an ads-* glob, so an unrelated skill such as ads-weather stays untouched.
• Expired claims that cannot be reverified are marked as failed current-evidence checks for that run.

Know what the skill needs before you run it

SKILL.md on its own does not contain the validator scripts. The checks depend on a Claude Ads repository checkout that includes the validator scripts and ownership manifests. Those files were not reviewed for this page, so confirm they exist before relying on any result.

The listing says the skill runs scripts and uses the network. The SKILL.md text does not list which network calls are made, so review the scripts in your checkout before running them on sensitive systems. The agent also needs access to the source files and filesystem it inspects. Checksum verification needs a trusted release channel.

• A Claude Ads repository checkout with validator scripts and ownership manifests.
• Read access to the source files and filesystem being checked.
• A trusted release channel for SHA-256 checksum verification, if you verify archives.
• Remote CI access if you need a verified remote CI result. The agent may not have it.