YARA rule 'agent_skill_credential_exfiltration_webhook': AI agent skill credential harvesting followed by webhook or external exfiltration [agent_skills]
- Category
- YARA Match
- Confidence
- 94% confidence
- Finding
The combination of creating webhooks to an external URL and handling secrets triggers a credible exfiltration concern. In this skill's context, directing users to register a webhook at a third-party domain can leak event metadata and create unintended outbound traffic; because the file discusses secrets and receiver implementation, the pattern is more dangerous than a generic webhook tutorial.
- Content
fixed example/evaluation convention. A real test needs a customer-controlled receiver. The example describes resource creation and read-back only; it does not claim that Baidu receives or processes Webhook notifications. For the receiver implementation, delivery test and signature verification in later sections, first replace the sample URL with the customer's actual controlled receiver.
bash curl --user-agent "$BAILIAN_MANAGED_AGENT_SKILL_UA" --max-time 30 -s -X POST "$BASE/webhook_endpoints" \ -H "Authorization: Bearer $API_KEY" -H "Content-Type: application/json" \ -d '{ "description": "Shibei E-commerce reconciliation-task notifier", "url": "https://www.baidu.com", "events": ["session.status_run_started", "session.status_idled", "deployment_run.started", "deployment_run.succeeded", "deployment_run.failed"] }'Illustrative creation response (key fields; the sample URL is updated to the evaluation convention):
json { "id": "wep_01M1G63CXHX6C7C1801RD2J