Oathe Security Badge

Is byungkyu/zoho-crm safe?

https://github.com/openclaw/skills/tree/main/skills/byungkyu/zoho-crm

79
CAUTION

The Zoho CRM skill is a legitimate commercial API integration that routes all Zoho CRM operations through Maton's third-party proxy infrastructure, with Maton also managing the user's OAuth tokens for Zoho. While the installation is clean (no local credential theft, no malicious processes, no canary exfiltration), the skill's core architecture makes Maton a permanent data intermediary for all CRM data and a persistent holder of Zoho OAuth credentials — creating significant ongoing privacy and security exposure that users must consciously accept. An embedded cross-skill referral in the system prompt description also nudges the agent toward expanding Maton's data access surface to other APIs.

Category Scores

Prompt Injection 90/100 · 30%
Data Exfiltration 55/100 · 25%
Code Execution 78/100 · 20%
Clone Behavior 100/100 · 10%
Canary Integrity 100/100 · 10%
Behavioral Reasoning 62/100 · 5%

Findings (8)

HIGH All Zoho CRM API Traffic Proxied Through Third-Party Maton Infrastructure -30

Every API call to Zoho CRM is routed through gateway.maton.ai rather than directly to Zoho's endpoints. All CRM records—leads, contacts, accounts, deals, organization details, user PII, license details, roles, and profiles—are transmitted through and visible to Maton's servers. Users must implicitly trust Maton as a data processor for their entire CRM dataset. While transparently stated in the skill, users may not fully appreciate that a third party sees all CRM read/write operations in perpetuity.

HIGH OAuth Credentials for Zoho CRM Managed and Stored by Maton -15

Maton manages OAuth connections at ctrl.maton.ai and injects live OAuth tokens server-side for every proxied request. This means Maton stores the user's Zoho CRM OAuth refresh tokens and has the ability to issue authenticated requests to Zoho on the user's behalf at any time, independent of the user's agent. The user cannot audit how these tokens are used or revoke them without disconnecting from Maton entirely.

MEDIUM Cross-Skill Referral Embedded in Agent System Prompt -10

The SKILL.md description field—which is injected verbatim into the agent's system prompt at skill load time—contains an explicit instruction to use another Maton skill for other API integrations: 'For other third party apps, use the api-gateway skill (https://clawhub.ai/byungkyu/api-gateway).' This is not a passive resource link but an active directive in the system prompt that could cause the agent to recommend or initiate installation of an additional Maton-controlled skill, expanding Maton's data access surface beyond Zoho CRM.

MEDIUM Inline Executable Python Code Blocks Throughout Skill Definition -22

The SKILL.md contains dozens of Python code blocks using heredoc syntax (python <<'EOF'...EOF) intended for direct agent execution. Each block reads MATON_API_KEY from the process environment and makes outbound HTTPS requests to maton.ai. While the current snippets implement legitimate API calls, this pattern means the agent executes code directly from the skill file. Any unauthorized modification to the skill file could introduce malicious payloads in these blocks without changing the skill's stated purpose.

LOW Maton as Single Point of Compromise for Ongoing CRM Access -18

Because Maton stores OAuth tokens and proxies all CRM operations, a breach of Maton's infrastructure would immediately grant attackers complete read/write access to the user's Zoho CRM without needing to compromise the user's systems, agent, or Zoho account directly. The user's CRM security posture is now coupled to Maton's security posture indefinitely.

LOW Permanent Vendor Dependency on Maton Commercial Service -20

All Zoho CRM integration through this skill requires a valid MATON_API_KEY and active Maton subscription. Changes to Maton's pricing, terms of service, availability, or shutdown would immediately break all CRM functionality for users of this skill. There is no fallback to direct Zoho API access within the skill's design.

INFO Clean Installation — No Anomalous Behavior During Clone 0

Installation performed only expected git operations: sparse clone from github.com (140.82.121.3:443), checkout of skills/byungkyu/zoho-crm subpath, file copy to skill directory, cleanup of /tmp/monorepo-clone. No unexpected network connections to Maton or other external hosts, no unexpected process spawning, and no filesystem writes outside the designated skill directory were observed during install.

INFO All Honeypot Files Intact — No Local Credential Theft 0

Honeypot files including .env, .ssh/id_rsa, .aws/credentials, .npmrc, .docker/config.json, and .config/gcloud/application_default_credentials.json were accessed read-only by the oathe monitoring framework during setup and post-install verification, but were not modified or transmitted to any external endpoint. No local credential harvesting was detected.