The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes npx skills commands to manage local installations and uses gh api to star the author's repository. The starring operation is performed silently without a user confirmation prompt.\n- [EXTERNAL_DOWNLOADS]: Fetches runtime scripts and configurations (such as catalog.cjs and session-check.cjs) from the author's repository on raw.githubusercontent.com. While the source is a well-known and trusted service, the behavior involving runtime script downloads is documented.\n- [REMOTE_CODE_EXECUTION]: The skill downloads script files and configures them to be executed by other AI agents during their session startup sequence.\n- [PERSISTENCE]: The skill modifies configuration files for Claude Code (settings.json), Codex (config.toml), and OpenCode (opencode.json) to install SessionStart hooks or plugins. These hooks are designed to execute the downloaded scripts from ~/.re0/ every time a new agent session begins.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingest data from the local environment, including installed skill lists and agent configurations, to drive its logic.\n
Ingestion points: Reads output from npx skills list, inspects the ~/.agents/skills/ directory, and parses agent-specific configuration files.\n
Boundary markers: Uses an internal catalog allow-list to validate and filter the skills it processes.\n
Capability inventory: Possesses capabilities to execute shell commands and write to user configuration directories.\n
Sanitization: The instructions require the agent to ignore unknown names and only act on specific cataloged names, reducing the risk of processing malicious inputs.

re0-upgrade