Skip to content

Repository files navigation

DTC Attribution Doctor

Your Marketing Doctor

Maintained by RTOAI.

简体中文

GitHub stars GitHub forks

If this skill is useful, star the repository to support the project and make it easier to find, or fork it to adapt the diagnostic workflows.

DTC Attribution Doctor is a lightweight, open Agent Skill for DTC ecommerce marketing diagnosis. It turns natural-language questions into focused analysis using Convbox first-party attribution data, then returns quantified findings, prioritized next steps, and daily, weekly, or custom-window reports.

Status: Preview. DTC Attribution Doctor is a starting point toward diagnosis-as-a-service. It is ready for evaluation and analyst-assisted workflows, but recommendations still require human review before execution.

Diagnose Before You Optimize

Dashboards show what changed. DTC Attribution Doctor is designed to help explain why it changed and what should be investigated next.

The working loop is:

Natural-language question
        -> automatic scenario routing
        -> first-party data retrieval
        -> diagnosis and comparison
        -> prioritized recommendations
        -> daily / weekly / custom report

The skill does not depend on users memorizing analysis-plan names. It detects intent, selects the relevant diagnostic workflow, keeps metric definitions consistent, and states when missing data prevents a reliable conclusion.

Capability Layers

Layer What it covers Representative outputs
Monitor and diagnose Growth anomalies, ROAS and CPA changes, attribution gaps, tracking health, creative and campaign performance, funnel loss, profit quality Root-cause findings, quantified gaps, ranked issues
Optimize Budget allocation, scaling decisions, bid strategy, audience efficiency, retargeting incrementality Prioritized recommendations with supporting evidence
Report Daily pulse, weekly review, monthly or custom-window analysis, role-specific reporting Decision-ready summaries for marketing teams and operators

Monitoring is host-triggered or scheduled batch analysis. It is not a real-time streaming alert service.

Why DTC Attribution Doctor

  • First-party data as the decision foundation: Convbox attribution metrics are the primary source of truth for diagnosis; Meta and Google self-reported metrics are comparison signals.
  • Automatic routing: users ask a business question in natural language and the skill selects the relevant analysis plan or plan chain.
  • Quantified attribution discrepancies: the skill can calculate and rank platform over-reporting by percentage and conversion-value gap instead of only saying that numbers disagree.
  • Portable across agent ecosystems: the same open skill is designed for Claude Code, Codex, OpenClaw, Hermes, WorkBuddy, GitHub Copilot, and other hosts that support SKILL.md workflows and HTTPS access.
  • Open and adaptable: customers and contributors can inspect the rules, adjust analysis plans, or propose new diagnostic scenarios.
  • Human-controlled execution: the skill recommends and explains; it does not silently change campaigns, bids, budgets, or storefronts.

Example Questions

My Meta ROAS fell last week. Can you diagnose the cause?
Why do Meta and Google report more conversions than our first-party data?
Which channels can safely receive more budget?
Where are customers dropping out of the onsite funnel?
Create a weekly paid-media review and prioritize what we should fix first.

Representative workflows include:

  • Growth health and campaign anomaly diagnosis
  • ROAS decline and CPA increase decomposition
  • Platform-reported versus first-party attribution discrepancy analysis
  • Tracking and match-rate health checks
  • Channel, campaign, audience, and creative performance analysis
  • Site funnel, landing-page reception, and add-to-cart diagnosis
  • Budget allocation, scaling, and bid-strategy guidance
  • Profit quality and retargeting incrementality analysis

Install

Ask your agent

Inspect https://github.com/RTOAI/dtc-attribution-doctor and its SKILL.md first.
Then install the dtc-attribution-doctor skill for my agent. After installation,
help me configure CONVBOX_API_KEY using secure environment settings. Never
print the key or write it into the repository, prompts, or logs.

One command

Install with the open Agent Skills CLI (Node.js 18+):

npx skills add RTOAI/dtc-attribution-doctor --skill dtc-attribution-doctor -g

The installer detects supported agents and lets the user select the target. To install manually for Codex:

git clone https://github.com/RTOAI/dtc-attribution-doctor.git \
  ~/.codex/skills/dtc-attribution-doctor

For another agent client, clone the repository into its documented skill directory or point the client at this repository root.

Requirements

  • A skill-capable AI agent that can read SKILL.md and make HTTPS requests
  • A Convbox account with a valid API key
  • Python 3.10+ and PyYAML 6.x only for the optional health checker

The hosted Convbox API is a separate service. The repository license does not grant API access or override the service terms associated with a Convbox account.

Configure the API key in the process environment. Do not add the real value to an .env file that may be committed.

macOS or Linux:

export CONVBOX_API_KEY="your-key"

Windows PowerShell:

$env:CONVBOX_API_KEY = "your-key"

Verify The Installation

Install the optional health-check dependency and validate local configuration:

python -m pip install -r requirements.txt
python utilities/config-health-check/config_health_check.py --config-only

Probe the documented API endpoints with a recent window:

python utilities/config-health-check/config_health_check.py \
  --recent-window 7 --strict

The checker never prints the API key. It reports OK, WARN, FAIL, or SKIP for each check.

How Analysis Is Routed

  1. SKILL.md identifies intent and required business context.
  2. functions.md selects the scenario, data interfaces, tier, and readiness.
  3. access.yaml defines exact API request and response fields.
  4. The matching file in plans/ defines data preparation, comparison, diagnosis, and recommended next steps.
  5. The agent reports the plans used and states data or business-input limits.

Repository Layout

.
|-- SKILL.md                 # Agent entry point, routing, and safety rules
|-- functions.md             # Data contracts and scenario catalog
|-- access.yaml              # Convbox API request and response dictionary
|-- plans/                   # Atomic analysis blueprints
|-- utilities/
|   `-- config-health-check/ # Credential, connectivity, and schema checker
|-- CONTRIBUTING.md
|-- SECURITY.md
`-- LICENSE

Data And Safety Boundaries

  • The skill consumes aggregated Convbox API responses; it does not perform customer-level ETL.
  • Platform passthrough data is for discrepancy analysis and must not be added directly to first-party attribution metrics.
  • Profit analysis requires configured cost data. Precise bid guidance requires a confirmed gross margin or an explicit user assumption.
  • Empty or unavailable data is reported as a limitation, never fabricated.
  • Outputs can vary by model. Validate high-impact recommendations with a human analyst and the relevant platform before execution.
  • Do not include customer exports, API responses, account IDs, or validation evidence in issues or pull requests.

Direction

The current release focuses on diagnosis, recommendations, and reporting. Future directions may include a unified severity framework, human-approved execution, persistent customer context, recommendation feedback loops, and shareable report artifacts. These are roadmap directions, not current release commitments.

Scenario readiness is authoritative in the Development status column of functions.md. A plan file alone does not imply production readiness.

Contributing And Security

See CONTRIBUTING.md before changing metric definitions, API fields, or recommendation thresholds. Report credentials, unauthorized data exposure, and authorization issues through SECURITY.md, not a public issue.

License

The repository is available under the MIT-0 license. Access to the hosted Convbox API is separate and requires an authorized account.

Convbox names and logos are trademarks of RTOAI. MIT-0 grants no trademark rights.

About

Open-source Agent Skill for diagnosing ROAS declines and attribution gaps for Shopify and Shopline DTC brands across Meta, Google Ads, Bing, TikTok powered by Convbox first-party attribution data.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages