The Agent Skills Directory

[PERSISTENCE]: The setup script modifies system shell configuration files (~/.bashrc and ~/.zshrc) to append a PATH export, ensuring that the installed CLI tools remain accessible across terminal sessions.
[REMOTE_CODE_EXECUTION]: The installation process fetches and executes remote shell scripts from external domains, including claude.ai and onecli.sh, via piped execution to shell interpreters (bash and /bin/sh).
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The gateway skill processes interactions with external services such as Gmail, GitHub, and Stripe as part of its core functionality (specified in payload/container/skills/onecli-gateway/SKILL.md).
Boundary markers: No explicit instruction delimiters or "ignore embedded instructions" warnings are provided for the data returned from these external sources.
Capability inventory: The skill enables broad network capabilities using curl, fetch, and other HTTP clients, and provides automated credential injection for outbound requests.
Sanitization: There is no evidence of sanitization or filtering of content ingested from external APIs before it enters the agent context.
[COMMAND_EXECUTION]: The skill uses child_process modules and shell scripts to manage Docker containers, install system binaries, and interact with the onecli and claude command-line interfaces.
[EXTERNAL_DOWNLOADS]: Fetches CLI binaries and release archives directly from GitHub and vendor-specific domains during the setup and upgrade phases.
[CREDENTIALS_UNSAFE]: The skill implements a migration logic to move sensitive credentials from plain-text .env files into a managed vault system; while this involves reading secrets, it is intended to improve long-term credential security.

add-onecli