concinno-skills-support 0.1.0

Customer support agent skills (Zendesk + Intercom) for Concinno — Databricks 2026 #1 enterprise landing use case.

concinno-skills-support

Customer-support agent skills for Concinno. Targets the Databricks 2026 #1 enterprise landing use case (customer service: finance 23% / retail 27%). Native Python REST wrappers, Apache-2.0 / MIT SDKs only — we deliberately avoid zenpy (GPL-3.0) to keep the Concinno ecosystem copy-left-free.

Status

MVP (0.1.0) — three one-shot tools:

Tool class	Platform	Underlying transport	Licence
ZendeskTicket	Zendesk	httpx (REST v2)	Apache-2.0
ZendeskSearch	Zendesk	httpx (REST v2)	Apache-2.0
IntercomMessage	Intercom	httpx (REST)	Apache-2.0

Both platforms are reached via Concinno's existing httpx main dependency; this package therefore adds zero additional runtime dependencies beyond Concinno itself.

Install

pip install concinno-skills-support

Credentials

Each platform's credentials live under well-known keys in the Concinno CredentialStore, which reads (in order):

1. Process runtime overrides via CredentialStore.set(...).
2. Env var CONCINNO_CRED_<UPPER_KEY>.
3. ~/.concinno/credentials.json.

Zendesk

Two supported auth shapes (OAuth is tried first):

Key	Env var
zendesk_subdomain	CONCINNO_CRED_ZENDESK_SUBDOMAIN
zendesk_email	CONCINNO_CRED_ZENDESK_EMAIL
zendesk_api_token	CONCINNO_CRED_ZENDESK_API_TOKEN
zendesk_oauth_token	CONCINNO_CRED_ZENDESK_OAUTH_TOKEN

API-token auth uses the official {email}/token:{api_token} Basic-auth format documented by Zendesk.

Intercom

Key	Env var
intercom_access_token	CONCINNO_CRED_INTERCOM_ACCESS_TOKEN

Example ~/.concinno/credentials.json:

{
  "zendesk_subdomain": "acme",
  "zendesk_email": "[e-mail]",
  "zendesk_api_token": "…",
  "intercom_access_token": "dG9rZW4="
}

If a token is missing the tool returns {"error": "no <service> credentials — set via CredentialStore / env"} rather than crashing.

Usage via Concinno ToolRegistry

When the consumer sets CONCINNO_LOAD_PLUGINS=1, the default registry auto-mounts every support tool:

import os
os.environ["CONCINNO_LOAD_PLUGINS"] = "1"

from concinno.tools.registry import get_default_registry

reg = get_default_registry()
assert {"ZendeskTicket", "ZendeskSearch", "IntercomMessage"} <= set(
    reg.list_deferred()
)

ticket = reg.get("ZendeskTicket")
ticket.call(action="list", status="open", limit=20)

Direct Python usage

from concinno_skills_support import (
    ZendeskTicket,
    ZendeskSearch,
    IntercomMessage,
)

ZendeskTicket().call(action="list", status="open", limit=50)
ZendeskTicket().call(
    action="create",
    subject="Login broken",
    comment="Customer reports 401 on sign-in",
    requester_email="[e-mail]",
    priority="urgent",
)
ZendeskSearch().call(
    action="search", query="status:open priority:urgent", limit=25
)
IntercomMessage().call(
    action="send",
    conversation_id="conv-42",
    body="We're on it — thanks for flagging.",
    admin_id="admin-7",
)

All tools return {"ok": True, ...} or a structured payload on success, or {"error": "..."} on failure — same shape as the other Concinno built-in tools. No exceptions escape the call() surface.

Concurrency

All three tools set is_concurrency_safe = False. Both Zendesk and Intercom enforce tenant-scoped rate limits, and ticket/conversation mutation is order-sensitive. The Concinno scheduler will honour this automatically.

Safety caps

• Zendesk search query ≤ 500 chars.
• Zendesk ticket comment body ≤ 50 000 chars.
• Intercom body ≤ 10 000 chars.

License

Apache-2.0. See LICENSE in the Concinno monorepo.