Back to skill

Security audit

Zapier

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent Zapier automation reference, but it needs review because it encourages broad automation, raw data forwarding/logging, and persistent account context without enough safeguards.

Install only if you are comfortable letting the agent help design and potentially operate broad Zapier workflows. Keep API keys in environment or managed secret fields, require explicit confirmation before enabling, deleting, sending messages, or calling arbitrary webhooks, and avoid logging or emailing raw payloads or putting personal data in URLs. Review and prune any ~/zapier/memory.md or main workspace memory entries it creates.

Vulnerability Patterns
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (3)

T09 · Insecure Skill Coding Practices

Error
Location
webhooks.md:259
Finding
Unrestricted Retention and Forwarding of Complete Webhook Payloads<![CDATA[ ## Vulnerability Details **File Location**: `webhooks.md:259-268` and `webhooks.md:319-326` **Vulnerability Type**: Sensitive-data exposure through excessive logging and email forwarding **Risk Level**: High ### Vulnerable Code ```text ### 3. Log Everything Action: Create Google Sheets Row Spreadsheet: Webhook Logs Data: Timestamp: {{zap_meta_human_now}} Zap ID: {{zap_meta_id}} Trigger Data: {{trigger.raw_body}} Response: {{webhook_response}} ``` ```text ### Webhook to Email Trigger: Catch Hook → Send Email: To: alerts@company.com Subject: New webhook: {{trigger.event}} Body: {{trigger.raw_body}} ``` ### Technical Analysis The Skill recommends storing the complete webhook request body and response in Google Sheets and forwarding the complete request body through email. Webhook payloads can contain personal information, payment or order details, access tokens, session identifiers, authorization headers, password-reset links, uploaded content, internal identifiers, and application-specific secrets. This behavior violates data-minimization and least-privilege principles. Operational diagnosis generally requires an event identifier, timestamp, result status, and sanitized error details—not unrestricted copies of all request and response data. Google Sheets and email may also have broader access, longer retention, forwarding, synchronization, and backup behavior than the source application. The instructions conflict with the Skill's security statement that connected applications receive only explicitly mapped data. Although `{{trigger.raw_body}}` is technically an explicit mapping, it indiscriminately maps every payload field and therefore defeats meaningful field-level minimization. ### Attack Path 1. An attacker or compromised upstream service submits a webhook containing sensitive values in the request body. 2. The Zap receives the payload through a Catch Hook or Catch Raw Hook. 3. The workflow follows the documented “Log Everyth ...[truncated 1072 chars]
Remediation
<![CDATA[ ## Remediation Suggestions 1. Replace “Log Everything” with an explicit allowlist of operational fields, such as: - Event ID - Event type - Receipt timestamp - Processing status - HTTP status code - Sanitized error category 2. Never log or email raw request bodies by default. 3. Recursively redact fields matching sensitive names, including `authorization`, `cookie`, `token`, `secret`, `password`, `api_key`, `session`, and `credential`. 4. Require explicit user confirmation before forwarding payload content to another service. 5. Separate diagnostic metadata from business payloads and apply role-based access controls. 6. Define retention periods and automatically delete old webhook logs. 7. Avoid email for sensitive diagnostic content; send a sanitized event identifier and provide a restricted audit-system link instead. 8. Add payload-size limits and schema validation before processing or recording webhook data. 9. Document that webhook responses may also contain secrets and must undergo the same redaction process. ]]>

T09 · Insecure Skill Coding Practices

Warning
Location
interfaces.md:50
Finding
Personal and Customer Information Embedded in Interface URLs<![CDATA[ ## Vulnerability Details **File Location**: `interfaces.md:50-53` and `interfaces.md:189-193` **Vulnerability Type**: Personal information exposure through URL query parameters **Risk Level**: Medium ### Vulnerable Code ```text **Pre-fill Values:** ``` ```text URL: https://yourform.zapier.app?name=John&email=john@example.com ``` ```text ### Variables in Chatbot Pre-fill context: ``` ```text URL: https://yourchatbot.zapier.app?customer_id=12345&name=John ``` ```text Bot can access: `{{customer_id}}`, `{{name}}` ``` ### Technical Analysis The documented examples place names, email addresses, and stable customer identifiers directly in URL query strings. Query parameters are commonly retained in browser history, reverse-proxy and web-server logs, monitoring platforms, analytics systems, bookmarks, copied links, screenshots, and support records. Depending on browser and server policy, they may also be included in outbound `Referer` headers. HTTPS protects the URL while it is in transit between endpoints, but it does not prevent storage or disclosure by the browser, endpoint infrastructure, logs, analytics, or users who copy the link. Stable customer identifiers can additionally enable account correlation or unauthorized context selection if an Interface trusts the parameter without server-side authorization. ### Attack Path 1. A workflow generates or distributes a personalized Interface URL containing an email address, name, or customer ID. 2. The recipient opens the URL. 3. The complete URL is retained by the browser and may be recorded by Zapier, proxies, analytics, security monitoring, or link-scanning systems. 4. The recipient shares the URL, uses a shared device, or navigates to another site under a permissive referrer policy. 5. Another party obtains the URL and learns the embedded personal information. 6. If the Interface uses `customer_id` as an authorization decision rather than untrusted display context, the party may modify the param ...[truncated 626 chars]
Remediation
<![CDATA[ ## Remediation Suggestions 1. Do not place names, email addresses, credentials, secrets, or stable customer identifiers in query strings. 2. Use short-lived, random, opaque context tokens that resolve to server-side state. 3. Bind each token to its intended user, Interface, purpose, and expiration time. 4. Make tokens single-use where practical and revoke them after submission. 5. Treat every URL parameter as untrusted input; never use `customer_id` alone for authentication or authorization. 6. Require authenticated server-side authorization before displaying customer-specific records. 7. Configure a restrictive referrer policy, preferably `Referrer-Policy: no-referrer`. 8. Redact query strings from access logs, analytics, error reports, and monitoring tools. 9. Update examples to use non-sensitive values, such as `?context_token=RANDOM_SHORT_LIVED_TOKEN`. ]]>

T02 · Agent Memory Poisoning

Note
Location
setup.md:11
Finding
Persistent Storage of User Stack and Activation Preferences<![CDATA[ ## Vulnerability Details **File Location**: `setup.md:11-17`, `setup.md:30-38`, and `memory-template.md:8-29` **Vulnerability Type**: Excessive persistent memory collection and cross-session behavioral influence **Risk Level**: Low ### Vulnerable Code ```text ### 1. First: Integration Ask early: - "Should I help whenever you mention automating tasks or connecting apps?" - "Want me to suggest automations proactively when I notice repetitive workflows?" Save integration preference to their main workspace memory. ``` ```text ## What to Save In ~/zapier/memory.md: - Apps they use (CRM, email, sheets, etc.) - Common workflows they need - Preference: simple vs advanced - Zapier plan (affects polling speed) - API access level ``` The persistent template also includes: ```markdown ## Account plan: free | starter | professional | team | company api_access: yes | no polling_speed: 15min | 2min | 1min ## Connected Apps <!-- Apps they use frequently --> - Google Sheets - Slack - Salesforce - (add as discovered) ## Active Zaps <!-- Zaps they've mentioned or created --> Format: Trigger App → Action App (purpose) ## Common Patterns <!-- Workflows they use repeatedly --> ## Preferences complexity: simple | advanced notifications: slack | email | both error_handling: alert | silent ``` ### Technical Analysis The Skill instructs the Agent to persist activation behavior in the main workspace memory and retain information about the user's applications, workflows, Zapier subscription, API access, active Zaps, and notification preferences. This creates cross-session state that can influence later Agent behavior and reveals potentially sensitive organizational architecture. Persistence is related to personalization and is disclosed, so it is not evidence of a covert backdoor. Nevertheless, writing activation preferences to general workspace memory exceeds the narrower requirement of keeping Zapier-specific preferences in `~/zapier/memory.md`. The instructions ...[truncated 1743 chars]
Remediation
<![CDATA[ ## Remediation Suggestions 1. Keep all Zapier-specific state inside the Skill-specific `~/zapier/` directory; do not modify general workspace memory. 2. Obtain explicit, informed consent before creating persistent memory. 3. Explain exactly which fields will be stored, why they are needed, and how long they will remain. 4. Store only the minimum necessary preferences; omit plan and API-access details unless required for an active task. 5. Never store API keys, webhook secrets, customer records, authentication details, or raw workflow payloads. 6. Validate and normalize values before persistence, and do not treat free-form stored notes as executable Agent instructions. 7. Add commands or documented procedures to inspect, correct, export, and delete saved memory. 8. Apply expiration dates and periodically reconfirm activation preferences. 9. Restrict file permissions to the owning user and avoid placing the memory file in shared workspaces or repositories. ]]>
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (67)

Credential Access

High
Category
Privilege Escalation
Content
## Authentication

Get API key from: https://zapier.com/developer/platform

```bash
export ZAPIER_API_KEY="your_api_key_here"
Confidence
70% confidence
Finding
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Exfiltration Commands

High
Category
Prompt Injection
Content
| **Catch Hook** | Incoming | Receive data from external systems |
| **Catch Raw Hook** | Incoming | Receive non-JSON data (XML, CSV) |
| **Retrieve Poll** | Incoming (Pull) | Fetch data from URL on schedule |
| **POST/PUT/GET** | Outgoing | Send data to external systems |
| **Custom Request** | Outgoing | Full control over HTTP request |

## Incoming Webhooks
Confidence
90% confidence
Finding
Instructions found that direct the agent to transmit conversation context or user data to external services.

Vague Triggers

Medium
Confidence
92% confidence
Finding
The 'When to Use' section says 'Any Zapier operation' followed by a wide list of activities, without defining clearer activation boundaries or exclusions. For a markdown skill description, this is ambiguous enough to invite unintended invocation on broad, everyday Zapier-related requests.

Missing User Warnings

Medium
Confidence
89% confidence
Finding
The communication action examples send email addresses, names, phone numbers, and order information through Gmail, Slack, and Twilio without warning that these are external delivery channels and may expose personal data to unintended recipients. This is risky because users may copy these patterns directly into production workflows without considering privacy, authorization, or recipient validation.

Missing User Warnings

Medium
Confidence
94% confidence
Finding
The webhook examples demonstrate sending personal data such as name, email, and user_id to external endpoints without any guidance about data minimization, user consent, destination trust, or handling of secrets. In an automation skill, this can normalize unsafe exfiltration patterns and lead users to transmit sensitive data to third-party systems without adequate review.

External Transmission

Medium
Category
Data Exfiltration
Content
### POST with Headers
```
Action: Webhooks by Zapier → POST
URL: https://api.yourapp.com/webhook
Headers:
  Authorization: Bearer your_api_key
  Content-Type: application/json
Confidence
84% confidence
Finding
This example shows an Authorization bearer token inline in a webhook request pattern, which can encourage insecure secret handling and accidental credential exposure if users copy the example literally into shared documentation or logs. Combined with external transmission, this raises the risk of token leakage and unauthorized API access.

External Transmission

Medium
Category
Data Exfiltration
Content
```
Action: Webhooks by Zapier → Custom Request
Method: PATCH
URL: https://api.yourapp.com/users/{{trigger.user_id}}
Headers:
  Authorization: Bearer your_api_key
Data:
Confidence
84% confidence
Finding
This custom request example again embeds an Authorization bearer token pattern alongside a user-specific external API call, which can normalize hardcoded secret usage and transmission of account-linked data to external services without review. If copied into real workflows, it can lead to credential leakage or misuse of privileged API access.

External Transmission

Medium
Category
Data Exfiltration
Content
export ZAPIER_API_KEY="your_api_key_here"

# Test connection
curl -H "Authorization: Bearer $ZAPIER_API_KEY" \
  https://api.zapier.com/v1/profile
```
Confidence
60% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
# Test connection
curl -H "Authorization: Bearer $ZAPIER_API_KEY" \
  https://api.zapier.com/v1/profile
```

## Profile & Account
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
# Test connection
curl -H "Authorization: Bearer $ZAPIER_API_KEY" \
  https://api.zapier.com/v1/profile
```

## Profile & Account
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
# Test connection
curl -H "Authorization: Bearer $ZAPIER_API_KEY" \
  https://api.zapier.com/v1/profile
```

## Profile & Account
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
# Test connection
curl -H "Authorization: Bearer $ZAPIER_API_KEY" \
  https://api.zapier.com/v1/profile
```

## Profile & Account
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
# Test connection
curl -H "Authorization: Bearer $ZAPIER_API_KEY" \
  https://api.zapier.com/v1/profile
```

## Profile & Account
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
# Test connection
curl -H "Authorization: Bearer $ZAPIER_API_KEY" \
  https://api.zapier.com/v1/profile
```

## Profile & Account
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
# Test connection
curl -H "Authorization: Bearer $ZAPIER_API_KEY" \
  https://api.zapier.com/v1/profile
```

## Profile & Account
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
# Test connection
curl -H "Authorization: Bearer $ZAPIER_API_KEY" \
  https://api.zapier.com/v1/profile
```

## Profile & Account
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
# Test connection
curl -H "Authorization: Bearer $ZAPIER_API_KEY" \
  https://api.zapier.com/v1/profile
```

## Profile & Account
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
# Test connection
curl -H "Authorization: Bearer $ZAPIER_API_KEY" \
  https://api.zapier.com/v1/profile
```

## Profile & Account
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
# Test connection
curl -H "Authorization: Bearer $ZAPIER_API_KEY" \
  https://api.zapier.com/v1/profile
```

## Profile & Account
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
# Test connection
curl -H "Authorization: Bearer $ZAPIER_API_KEY" \
  https://api.zapier.com/v1/profile
```

## Profile & Account
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
# Test connection
curl -H "Authorization: Bearer $ZAPIER_API_KEY" \
  https://api.zapier.com/v1/profile
```

## Profile & Account
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
# Test connection
curl -H "Authorization: Bearer $ZAPIER_API_KEY" \
  https://api.zapier.com/v1/profile
```

## Profile & Account
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
# Test connection
curl -H "Authorization: Bearer $ZAPIER_API_KEY" \
  https://api.zapier.com/v1/profile
```

## Profile & Account
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
# Test connection
curl -H "Authorization: Bearer $ZAPIER_API_KEY" \
  https://api.zapier.com/v1/profile
```

## Profile & Account
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
# Test connection
curl -H "Authorization: Bearer $ZAPIER_API_KEY" \
  https://api.zapier.com/v1/profile
```

## Profile & Account
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Static analysis

Detected: suspicious.prompt_injection_instructions

Prompt-injection style instruction pattern detected.

Warn
Code
suspicious.prompt_injection_instructions
Location
interfaces.md:159