yysun/crm-chat/rlp-crm

RLP CRM

This skill answers CRM questions against the live CRM API. The source of truth is scripts/data-tool.js; do not use web browsing, web_fetch, curl, or hand-built /api/data/... calls.

The script owns the CRM read boundary. It builds only supported read operations, loads its own configuration, applies environment-owned auth headers, calls the CRM API, and prints a JSON response. The agent must not search for, read, print, or infer .env, API_BASE_URL, API_PAT, bearer tokens, authorization headers, or security context values.

Required Helper

Use the bundled helper for every CRM data read:

node scripts/data-tool.js list
node scripts/data-tool.js search --q "alex"
node scripts/data-tool.js account --id 123
node scripts/data-tool.js actions --date 2026-05-20

If running from the workspace root instead of this skill directory, use:

node .agents/skills/rlp-crm/scripts/data-tool.js search --q "alex"

The helper reads API_BASE_URL and API_PAT itself from the project-root environment. Do not inspect or pass those values. If configuration is missing, report the helper error and stop.

For debugging, payload prints the validated request without calling the API:

node scripts/data-tool.js payload search --q "alex"

If the helper rejects a request, fix the operation or inputs. Do not bypass the helper for routes it supports.

Response Handling

The helper returns a single JSON object:

• ok, status, statusText, url, headers
• body, parsed from JSON when possible

If ok is false, report the failed status and do not invent missing data. If the response body shape is unexpected, say what came back and adjust the next call.

Supported Operations

• who: calls GET /api/data/who.
• search --q <text>: calls GET /api/data/contacts/searchAll?q=....
• account --id <id>: calls GET /api/data/accounts/:id.
• account-contacts --id <id>: calls GET /api/data/accounts/:id/contacts.
• account-notes --id <id>: calls GET /api/data/accounts/:id/notes.
• actions [--date YYYY-MM-DD]: calls GET /api/data/actions?date=...; date

defaults to today when omitted.

• contact --id <id>: calls GET /api/data/contacts/:id.
• contact-notes --id <id>: calls GET /api/data/contacts/:id/notes.

The helper also accepts repeatable non-auth --header Name=Value options where the API supports them.

Operating Judgment

Use search --q <text> for ambiguous name searches because it searches both contacts and accounts.

For account or company questions, use the account ID if provided. Otherwise search by name first, choose the matching account, then load the account detail route.

For contact questions, use the contact ID if provided. Otherwise search by name first, choose the matching contact, then load the contact detail route.

For notes questions by name, search first to identify the account ID or contact ID, then call the matching notes route.

For status, triage, "what's going on", next-move planning, and action questions, default to today's actions unless the user gives a different date. For account/contact versions of those questions, fetch by date only and match the account/contact inside the action text.

Load detail routes before making claims about a single record. For summaries, cite which record IDs and routes supported the conclusion.

Do not claim data exists unless it came from scripts/data-tool.js.