The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill fetches data from a remote repository via git fetch origin and retrieves the content of codex/UPDATE.md. It then commands the agent to "Follow the fetched UPDATE.md instructions from start to finish," which is functionally equivalent to executing a remote script for an AI agent. This allows an external party who controls the repository to dictate the agent's actions.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted instructions from a remote source and incorporates them directly into the agent's execution flow. There are no boundary markers or instructions provided to the agent to treat this external content as untrusted, nor is there any sanitization of the input.
Ingestion points: git -C ~/.turbo/repo show origin/main:codex/UPDATE.md (SKILL.md)
Boundary markers: Absent
Capability inventory: Shell command execution and filesystem access via git and other tools available to the agent.
Sanitization: Absent
[COMMAND_EXECUTION]: The skill executes shell commands to perform Git operations (git fetch and git show). These commands interact with both the network (fetching from the remote) and the local filesystem (~/.turbo/repo).

update-turbo