The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on executing local shell scripts (e.g., scripts/assert-mirror-current.sh) and CLI tools (relayfile, agent-relay) to perform workspace management and health checks.
Evidence: scripts/assert-mirror-current.sh uses subprocess.run in Python to invoke relayfile commands.
[CREDENTIALS_UNSAFE]: The skill uses environment variables (RELAY_ENROLLMENT_TOKEN, RELAY_AGENT_TOKEN) to manage authentication tokens and explicitly acknowledges a risk where tokens passed as command-line arguments are visible in process listings.
Evidence: SKILL.md documents a 'Credential hazard' where tokens are readable via ps auxww and suggests rotating tokens seen in process listings.
[INDIRECT_PROMPT_INJECTION]: The skill establishes a data ingestion pipeline from external providers into the agent's working environment, which could contain malicious instructions.
Ingestion points: Data is fetched from the cloud via relayfile read and relayfile tree, then presented to the agent in a local mount directory.
Boundary markers: The skill provides a suggested task prompt that identifies the environment as a 'live Relayfile mount' and provides usage constraints, which helps the agent distinguish system instructions from workspace data.
Capability inventory: The agent has the ability to read and write files within the mounted tree and execute relayfile CLI commands.
Sanitization: The provided scripts perform basic path validation but do not sanitize the content of the files read from the cloud providers.

multi-host-live-mount