The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documentation and provided scripts (assert-mirror-current.sh, assert-known-true-now.sh, assert-cross-host-write-visible.sh) execute the relayfile and agent-relay CLI tools to manage workspace mounts, join fleets, and perform data integrity checks.
[CREDENTIALS_UNSAFE]: The skill handles workspace enrollment tokens and agent tokens. It promotes secure practices by using shell 'read -s' to prevent tokens from appearing in shell history and explicitly warns about the danger of passing secrets as command-line arguments which could be exposed in process listings (e.g., via ps auxww).
[DATA_EXFILTRATION]: The assert-mirror-current.sh script performs an integrity check by reading the contents of files in the local mount and comparing them against the version stored in the cloud via the relayfile CLI. This operation is essential for the skill's purpose of verifying that local mirrors are not stale and utilizes the vendor's own API.
[DYNAMIC_EXECUTION]: The shell scripts utilize inline Python code blocks to implement complex directory traversal and file comparison logic. This code is contained within the skill's scripts and is executed locally to verify mount health without external code injection.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the placement of agents in environments that ingest data from external providers (GitHub, Linear, Notion). While the skill focuses on the mounting mechanism, the resulting environment creates a surface where external content is processed by the agent, which is a standard risk for agents operating on external data.

multi-host-live-mount