Dark Social Attribution Analyst
Turn untraceable sharing into clearly labeled, auditable attribution estimates.
The skill shows a least-privilege posture: it analyzes user-provided data, forbids posting, pixel injection, and DM automation, treats pasted exports as untrusted, and asks before saving memory. Writes to memory, registry proposals, and self-reported attribution data still depend on host and user controls; isolation, revocation, and sensitive-data handling are not fully demonstrated in the supplied file, so 4 points are deducted.
Triggers, inputs, outputs, completion criteria, failure boundaries, and handoff routes are comparatively explicit, and derived numbers must be labeled Estimated/proxy. Referenced files and pageviews.py were not independently reviewed here, and there are no skill-specific tests or execution feedback; the static cap yields 9 points, with 1 point deducted.
Audience, use cases, non-fit boundaries, and Chinese form/WeChat paths are clearly stated, with pasted GA4/GSC exports supported. The skill relies on user-supplied data, while mainland-China reachability and fallback paths for Wikipedia/pageviews and GA4/GSC are not sufficiently specified; trigger precision is also mostly textual, so 5 points are deducted.
The document is well layered and includes Quick Start, contract, sources, procedures, save rules, references, version, Apache-2.0 licensing, and limitations. Maintenance ownership, changelog/update path, and governance for the unverified publisher are unclear, and installation/troubleshooting details for referenced materials are not fully shown, so 5 points are deducted.
The expected pack is concrete and directly usable: UTM hygiene, self-report field design, GA4 decomposition, branded-search proxy, and method documentation. Estimated/proxy labeling reduces misuse. However, the core result is heuristic, requires complete user exports and manual review, and lacks verified representative outputs; the static cap yields 6 points, with 1 point deducted.
The skill names sources, denominators, and labels and requires separation of Measured, User-provided, and Estimated evidence, providing some auditability. No skill-specific execution record, test suite, or independent corroboration is supplied; repository-level tests alone do not reproduce this key path, so it receives 3 points.
- GA4 direct-traffic decomposition, mobile skew, push correlation, and branded-search lift are proxies or heuristics, not incrementality or ROI proof.
- The self-reported source field may process personal data; confirm notice, retention, access control, and data minimization before deployment.
- pageviews.py, GSC/GA4 procedures, and referenced protocol files were not separately verified in this static review; prepare user exports or alternatives if mainland-China access is unavailable.
- Do not write memory or registry state automatically; preserve the confirmation and sole-writer/proposal constraints.
What does this skill do, and when should you use it?
Dark Social Attributor is an Observe-phase social skill in the Aaron Marketing Skills repository for analyzing direct traffic with missing referrers. It combines GA4 direct-traffic heuristics, Google Search Console branded-query trends, Wikipedia pageviews, share-link UTM hygiene, and self-reported attribution design. Derived figures must be labeled Estimated or proxy and are never presented as Measured. It fits teams investigating the impact of DMs, group chats, newsletter forwards, and copied URLs without reliable click data.
It reads user-provided GA4 landing-page or channel exports, branded-query series from GSC, share-surface and form inventories, and read-only channel dossiers. It decomposes direct traffic using deep-URL directs, mobile-app skew, and time-boxed correlation with private pushes, labeling each heuristic as Estimated/proxy. It can run scripts/connectors/pageviews.py to obtain a Wikipedia attention series as an external control. It produces a share-link and UTM hygiene specification, replaces an existing low-value form field with a free-text “How did you hear about us?” design and later coding plan, and packages the findings into a declared-method document and handoff summary.
- A growth team sees unexplained growth in GA4 direct traffic and wants to estimate how much may come from copied or privately shared links.
- A content team needs a stable short-link and UTM taxonomy for blog, documentation, newsletter, and community share surfaces.
- A product team wants to add self-reported acquisition data without adding another signup-form field.
- A social lead needs a proxy read on group chats, newsletter forwards, or private community pushes when platform click data is unavailable.
- An analytics team wants to compare branded-search trends and Wikipedia pageviews with a social activity calendar.
What are this skill's strengths and limitations?
- Maintains a strict distinction between Measured, User-provided, Estimated, and proxy evidence.
- Covers instrumentation, form-based attribution, GA4 decomposition, and branded-search proxy analysis in one workflow.
- Keeps organic dark-social analysis separate from paid-channel attribution reconciliation.
- Works from user-provided GA4/GSC data and a keyless Wikipedia pageview connector under the repository’s Tier 1 model.
- Sets explicit boundaries around closed platforms and prohibits tracking injection and automation.
- It cannot directly measure traffic from DMs, group chats, forwards, or copied URLs; it can only estimate plausibility.
- Closed-platform analysis depends on user-supplied native analytics exports.
- A complete read requires GA4/GSC exports, share-surface details, and current form fields.
- The Wikipedia proxy applies only where a relevant brand page exists.
- The source provides no independent test suite or accuracy benchmark for this individual skill.
How do you install this skill?
The repository README documents generic installation with npx skills add aaron-he-zhu/aaron-marketing-skills -s dark-social-attributor, or by cloning https://github.com/aaron-he-zhu/aaron-marketing-skills. Claude Code users can also run /plugin marketplace add aaron-he-zhu/aaron-marketing-skills, followed by /plugin install aaron-marketing@aaron. The standalone skill file is social/observe/dark-social-attributor/SKILL.md. Host-specific loading instructions beyond these documented options are not provided.
How do you use this skill?
Provide the compatible Agent Skills host with GA4/GSC exports, a share-surface inventory, and current form fields. A supported trigger is: “Decompose our GA4 direct traffic for the last 90 days—here is the landing-page export. How much is plausibly dark social?” The skill returns UTM/share-link guidance, a replacement self-reported attribution field, labeled direct-traffic estimates, a branded-search-lift proxy, and a declared-method document. It asks for confirmation before saving results for future sessions.
How does this skill compare with similar options?
This is not attribution-reconciler, which handles paid-platform conversion claims and analytics reconciliation, nor social-measurement-loop, which owns the metric dictionary and write-back loop. Dark Social Attributor declares the estimation method and specifies the instrumentation first.