Retail Agent Setup

Onboarding wizard for retail digital employee agents — guides businesses through a 12-step setup to configure a fully operational AI store assistant. Use whe...

Security Scan

Skill appears to implement a reasonable retail onboarding flow, but there are notable operational and transparency gaps (undocumented runtime dependencies, no install spec, and references to credential env vars) that warrant caution and code review before use.

Purpose & Capability

Name, description, and runtime instructions align with a retail onboarding wizard. Included parsing and test-generation scripts (product/policy parsing, scoring, test-case generation) fit the stated purpose. Minor inconsistency: SKILL.md shows use of environment variables (example: INVENTORY_API_KEY, channel credential names) and external webhook endpoints as part of channel setup, but the registry metadata lists no required env vars — this is reasonable for optional/conditional connectors but should be explicit in the manifest.

Instruction Scope

Instructions are detailed and scoped to onboarding tasks: collecting system inventory, importing files (CSV/XLSX/PDF/DOC/images), running parse scripts, configuring channels and escalation, and saving artifacts to agent memory. The skill explicitly warns not to import customer PII into the knowledge base and recommends using API queries for PII — that is good. Still, the agent is instructed to accept file uploads and run parsing scripts on those files; users should confirm where uploaded files and parsed outputs are stored and who/what can access them.

Install Mechanism

No install spec is provided even though the package includes Python scripts and a scripts/requirements.txt. The developer notes require libraries like pandas and pdfplumber in _dev/decisions.md and requirements.txt, but there is no automated install step (pip, brew, or similar). That means runtime failures or manual dependency installation will be required. The absence of an install mechanism is an operational risk and also increases the chance someone will run scripts in an environment without inspecting them first.

Credentials

The registry shows no required environment variables, and the skill does not demand unrelated credentials. However SKILL.md and references use example env var names for connectors (e.g., INVENTORY_API_KEY, WECOM_* , WECHAT_*), and channel setup instructs users to provide API credentials and webhook URLs. This is proportionate to purpose but the manifest should declare optional env vars or clearly document which variables are needed when certain channels/skills are enabled.

Persistence & Privilege

The skill stores onboarding artifacts in the agent's memory under a named key (retail_setup_state) and references saving config objects (role_config, skills_config, etc.). always:false (not force-included) and there is no indication it modifies other skills or system-wide agent settings. Persisting state to agent memory is coherent with resumable onboarding.

What to consider before installing

This skill is consistent with a retail onboarding wizard, but take these precautions before installing or running it: - Review the included Python scripts (scripts/parse_*.py, gen_test_cases.py, score_knowledge.py) before execution to ensure they do not transmit data externally or perform unexpected network activity. - The bundle includes scripts/requirements.txt but no install spec — you or your operator must provision a Python runtime and install dependencies (e.g., pandas, pdfplumber) before using the parsing features. Prefer running in a sandboxed environment first. - Do not upload raw customer PII to the knowledge base. The SKILL.md warns about this; follow it: use API connections for PII or anonymize data prior to import. - Expect to provide channel credentials (WeCom, WeChat MP, WhatsApp/BSP, inventory API keys) during setup if you enable those connectors. The skill references env var names and webhook URLs; validate where those credentials will be stored and who can access them. - Verify where agent memory and any temporary files (parsed JSON, OCR outputs) are stored and how long they persist. If your organization has data residency or retention policies, confirm compatibility. - If you plan to enable autonomous model invocation or allow the agent to run tests/configuration unattended, consider a staged rollout: test with non-sensitive sample data, confirm the test report and guardrails behave as expected, and only then run against production data. Overall: the skill is plausibly benign in intent, but the missing install/dependency instructions and the presence of executable parsing scripts mean you should inspect and test the code and runtime environment before trusting it with production data or credentials.

Retail Agent Setup — Onboarding Wizard

Overview

This skill transforms a blank OpenClaw agent into a fully configured retail digital employee tailored to a specific store or chain. Each step produces a concrete artifact that persists in the agent's memory, making the setup cumulative and resumable.

Setup takes 20–40 minutes end-to-end. Each step can be paused and resumed. Run retail agent setup or 数字员工配置 to start or continue.

---

Execution Protocol

• Run steps in order — each step depends on outputs from the previous
• Pause after each step — show the artifact, ask "Confirm and continue?" before proceeding
• Resumable — if a step was previously completed, show its saved output and ask whether to redo or skip
• Save state — write each step's output to agent memory before moving to the next
• Zero-config entry — if the user just says "set up my retail agent," start at Step 1

---

The 12 Steps

Step 01 — System Inventory

"What retail systems are you currently using?"

Identify the store's existing tech stack across 5 categories: POS, ERP/WMS, CRM/membership, e-commerce platforms, and supply chain tools.

Map each system to its API availability (real-time / batch / none). Reference: step-01-systems.md Artifact: System inventory card + API availability matrix

---

Step 02 — Data Infrastructure Assessment

"Where does your data live, and what format is it in?"

Evaluate data across 6 dimensions: products, inventory, sales, staff, customers, and policy docs. Score completeness and freshness. Prioritize what to connect first. Reference: step-02-data-infra.md Artifact: Data map + connection priority list

---

Step 03 — Data Import & Auto-Structuring

"Send me your data — I'll organize it into a format the agent can use."

Accept uploads (Excel/CSV/PDF/Word/image), API connections, or pasted text. Auto-parse into structured knowledge base entries. Flag gaps and prompt to fill them. Script: scripts/parse_products.py — Excel/CSV → structured JSON Script: scripts/parse_policy.py — PDF/Word → rule tree Script: scripts/score_knowledge.py — completeness scoring Reference: step-03-data-import.md Artifact: Structured knowledge base + completeness score (0–100)

---

Step 04 — Role Selection

"What role should this digital employee play?"

Choose from 6 preset roles or define a custom role. Each role activates a specific skill bundle and response style. One agent = one primary role (multi-role is advanced config). Reference: step-04-role-select.md Artifact: Role definition file + activated skill bundle list

---

Step 05 — Skills Configuration

"Which capabilities should this agent have?"

Review recommended skills for the chosen role. Toggle on/off. Configure each enabled skill (thresholds, data sources, escalation rules). Reference: step-05-skills-config.md Artifact: skills-config.json — active skills with their parameters

---

Step 06 — Knowledge Base Validation

"Let me test what your agent knows."

Auto-generate 10 test questions covering products, inventory, policies, and recommendations. Run them against the knowledge base. Flag failures. Guide the user to fill gaps. Script: scripts/gen_test_cases.py — generate test questions by vertical Script: scripts/score_knowledge.py — run and score responses Reference: step-06-knowledge.md Artifact: Knowledge base score + gap report

---

Step 07 — Digital Employee Persona

"Give your digital employee a name and personality."

Configure: name, personality type, tone, reply style, customer address form, brand keywords. Generate 3 sample dialogues for preview. Confirm before saving. Reference: step-07-persona.md Artifact: persona-config.json + 3 preview dialogues

---

Step 08 — Channel Integration

"How will staff and customers reach this agent?"

Select and configure delivery channels: WeCom (企业微信), WeChat MP/Mini Program, Lark (飞书), Web kiosk UI, WhatsApp, or SMS/IVR. Each channel has a dedicated setup guide with step-by-step auth instructions. Reference: step-08-channels.md Artifact: Channel connection status + test message confirmation

---

Step 09 — Permissions & Escalation

"What can the agent decide alone, and what needs a human?"

Define 4-level permission matrix: L0 auto-handle, L1 suggest+confirm, L2 submit for approval, L3 force escalate to human. Set escalation targets and on-call schedules. Reference: step-09-permissions.md Artifact: permissions-matrix.json + escalation routing config

---

Step 10 — Pre-Launch Testing

"Let's run real-scenario tests before going live."

Run a full scenario test suite based on the store's vertical and configured skills. Score readiness 0–100. Must reach 80+ to proceed to launch. Script: scripts/gen_test_cases.py Reference: step-10-test.md Artifact: Test report + launch-readiness score

---

Step 11 — Launch & Handoff

"You're ready. Let's go live."

Activate the agent on all configured channels. Generate staff onboarding card (one-pager). Send welcome message. Schedule first check-in reminder (7 days out). Reference: step-11-handoff.md Artifact: Staff guide PDF + activation confirmation

---

Step 12 — Continuous Improvement

"Going live is the beginning, not the end."

Set up weekly unanswered-question digests and monthly usage reports. Configure knowledge-gap alerts. Schedule quarterly persona review. Reference: step-12-iterate.md Artifact: Cron jobs for digest + alert thresholds set

---

State Management

Track onboarding progress in agent memory under key retail_setup_state:

{
  "version": "1.0",
  "started_at": "<ISO timestamp>",
  "completed_steps": [1, 2, 3],
  "current_step": 4,
  "artifacts": {
    "systems": { ... },
    "data_map": { ... },
    "knowledge_base": { ... },
    "role": "...",
    "skills_config": { ... },
    "persona": { ... },
    "channels": [ ... ],
    "permissions": { ... }
  }
}

On any new message, check this state first. If setup is incomplete, offer to resume.

---

Supported Retail Verticals

Apparel · Footwear · Beauty & Skincare · Consumer Electronics · Home & Furniture · Maternal & Infant · Convenience Store · Supermarket · Specialty Food · Jewelry · Sporting Goods · Books & Stationery · Pet Supplies · Pharmacy · Toy & Hobby

For verticals not listed, use "General Retail" defaults and customize in Step 4.