The Agent Skills Directory

Remote Code Execution (HIGH): The skill promotes a highly dangerous installation pattern: curl -sSL https://canifi.com/skills/datadog/install.sh | bash. This executes unverified remote scripts with the user's shell privileges.
External Downloads (HIGH): The skill relies on scripts and tools hosted at canifi.com, which is not a trusted source according to security standards. This includes the main skill installer and the canifi-env management tool.
Unsafe Credentials (HIGH): The 'Option 2: Environment Variables' setup instructs users to store SERVICE_PASSWORD in plaintext via canifi-env. While the skill claims these are only local, they are ingested by the LLM agent to perform browser automation, posing a significant risk of credential exposure or exfiltration if the agent's context is compromised.
Indirect Prompt Injection (LOW): The skill is designed to ingest and process untrusted external data from Datadog logs and APM traces.
Ingestion points: Data is retrieved from app.datadoghq.com (Logs, APM, Dashboards).
Boundary markers: No delimiters or instructions are provided to the agent to ignore potentially malicious content within logs.
Capability inventory: The skill utilizes Playwright for browser automation and has access to the local terminal via canifi-env.
Sanitization: There is no evidence of sanitization or filtering of the content retrieved from Datadog before it is processed by the agent.

datadog