Google analytics
the governed runtime for agent skill workflows, off the leash but on the record
npx -y skills add runxhq/runx --skill google-analyticsAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
What its author says it does
Copied from the file, not written here
Read and normalize GA4 properties, metadata, standard reports, and realtime reports through any compatible Runx connector, preserving the measurement caveats an operator needs before turning analytics into action.
SKILL.md
5.4 KB, as published. Nobody here has run it
Google Analytics
Use this skill for bounded GA4 evidence: finding accessible properties, checking the dimension and metric catalog, measuring acquisition and on-site outcomes, or reading a short realtime window after a deployment or campaign.
The portable skill owns the measurement procedure and evidence contract. It does not own OAuth, credentials, tenants, connector selection, or provider routing. Operators may bring a local, self-hosted, third-party, or Runx-hosted connector. Any of those is compatible when it implements the operations and scopes below. Never place a connection id, credential handle, provider-config key, service URL, or tenant id in a skill input, fixture, or evidence packet; those bindings belong in the operator's Runx grant and connector configuration.
Read the right thing
Use properties when the property is unknown. Use metadata before composing
an unfamiliar report: GA4 dimension and metric compatibility is part of the
query contract, not a formatting concern. Use report for date-bounded
analysis and realtime only for a current operational pulse.
normalize-report accepts a GA4-shaped result already supplied by a trusted
import or connector test. It performs the same semantic checks, but its output
says provider_status: not_called and source_status: supplied_result. A
supplied packet is useful evidence with external provenance; it is never proof
that Runx reached Google.
A report should name the exact property, date ranges, ordered dimensions, ordered
metrics, filters, ordering, currency, row limit, and offset needed for the
decision. Keep it narrow. Wide reports are more likely to combine incompatible
fields, hit quotas, hide detail in an (other) row, or give an agent an
attractive but irrelevant aggregate.
The normalized packet maps each ordered value to its header name, retains row count and pagination, and carries timezone, currency, quota, and response metadata. Header drift is a hard failure because a row cannot be interpreted safely when the requested and returned fields differ.
Measurement caveats
Google Analytics is behavior and outcome evidence, not search-ranking evidence.
The Organic Search channel can include search engines other than Google and
depends on the property's channel definitions. Landing-page sessions and key
events help quantify outcomes, but they do not explain why visibility changed.
Join GA4 with Search Console on a deliberate page and date boundary, then inspect
content, releases, attribution settings, and business context before making a
causal claim.
Treat thresholding, sampling metadata, schema restrictions, and
data_loss_from_other_row as material. The skill returns
decision: usable_with_caveats when any is present. Do not silently compare a
thresholded segment with an unthresholded total, add rows across overlapping date
ranges, or treat a paginated page as the complete property.
Timezone, currency, reporting identity, key-event configuration, consent mode, and attribution settings can change interpretation without changing the query. Carry those assumptions into any comparison. Realtime data is provisional and suited to detection, not settled performance reporting.
Connector capability contract
A compatible connector exposes provider google-analytics with:
properties.listunderproperties.read, projectingproperties.metadata.getundermetadata.read, projecting the property, dimensions, metrics, and fetch time.reports.rununderreports.read, projecting property, request, dimension and metric headers, rows, row count, pagination, response metadata, property quota, and fetch time.reports.run_realtimeunderreports.realtime.read, using the same projected report shape and an explicit realtime request.
The connector decides how credentials are acquired and refreshed. It must enforce the caller's grant, bind the requested property, preserve Google errors, project only declared fields, and keep secret material out of results and receipts. This package is read-only; analytics administration, audience changes, event configuration, and deletion requests require separate governed capabilities.
Operator handoff
A sealed report packet is ready for review or for a higher-level
organic-growth comparison. Preserve the evidence digest and the exact query
when deriving trends. A useful organic acquisition slice commonly includes a
landing-page dimension, a channel or source dimension, and only the outcome
metrics needed for the decision, such as sessions, engaged sessions, key events,
or revenue.
Stop when the property is ambiguous, requested fields are incompatible, response headers drift, pagination is incomplete for a whole-property claim, privacy metadata undermines the comparison, timezone or currency is unknown where it matters, the supplied result lacks provenance, or the configured connector lacks the exact capability.
The deterministic module exists for GA4 domain semantics that generic Runx tools cannot infer: binding ordered headers to values, validating report identity, parsing numeric metrics, and carrying privacy and completeness caveats. Runx owns provider execution, grants, digesting, projection, and receipts.