The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructions manage authentication credentials by storing and reading tokens from local paths such as ~/.relayfile/credentials.json and ~/.relayfile/cloud-credentials.json.
[COMMAND_EXECUTION]: The instructions rely on various shell commands (relayfile, curl, jq, daytona) to configure the environment, start background mount daemons, and diagnose failures.
[DATA_EXFILTRATION]: Provides examples of reading local authentication tokens using jq and transmitting them to the agentrelay.com domain via curl headers to facilitate session connectivity.
[PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by mounting untrusted content from external providers (e.g., Slack, Notion) into the agent's filesystem.
Ingestion points: Files located in the ~/relayfile-mount/ directory, which mirrors data from third-party services.
Boundary markers: There are no instructions for implementing boundary markers or "ignore" directives for the data synced from external providers.
Capability inventory: Includes the relayfile CLI toolset, network operations via curl, and shell command execution via tools like daytona.
Sanitization: No specific filtering or sanitization steps are documented for the content being read from external providers.

setting-up-relayfile