Security Audit
Replicate Automation
github.com/ComposioHQ/awesome-claude-skillsTrust Assessment
Replicate Automation received a trust score of 82/100, placing it in the Mostly Trusted category. This skill has passed most security checks with only minor considerations noted.
SkillShield's automated analysis identified 2 findings: 0 critical, 1 high, 1 medium, and 0 low severity. Key findings include Arbitrary Webhook URL for Prediction Results, Base64 File Upload Capability.
The analysis covered 4 layers: Manifest Analysis, Static Code Analysis, Dependency Graph, LLM Behavioral Safety. All layers scored 70 or above, reflecting consistent security practices.
Last analyzed on February 20, 2026 (commit 27904475). SkillShield performs automated 4-layer security analysis on AI skills and MCP servers.
Layer Breakdown
Behavioral Risk Signals
Security Findings2
| Severity | Finding | Layer | Location | |
|---|---|---|---|---|
| HIGH | Arbitrary Webhook URL for Prediction Results The `REPLICATE_MODELS_PREDICTIONS_CREATE` tool allows specifying an arbitrary `webhook` URL. A malicious prompt could instruct the agent to set this webhook to an attacker-controlled server, thereby exfiltrating the results of any prediction, which may include sensitive data provided in the `input` parameter. Implement strict allow-listing for webhook URLs, or require explicit user confirmation for external URLs. Restrict the LLM's ability to set arbitrary URLs for this parameter without explicit user consent or a predefined safe list. | LLM | SKILL.md:46 | |
| MEDIUM | Base64 File Upload Capability The `REPLICATE_CREATE_FILE` tool accepts `base64-encoded file content`. If the AI agent has access to other tools that can read local files (e.g., a `read_file` tool), a malicious prompt could instruct the agent to read sensitive local files, base64-encode their content, and then upload them to Replicate. While Replicate is a legitimate service, this capability could be abused for data exfiltration if the Replicate account is compromised or if the data is not intended to leave the local environment. Implement strict controls on what files the agent can read and upload. Require explicit user confirmation for uploading content that originates from local files, especially sensitive ones. Consider sandboxing file access for the agent. | LLM | SKILL.md:61 |
Scan History
Embed Code
[](https://skillshield.io/report/ac43d78e714936e9)
Powered by SkillShield