The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of CLI tools including ao, gc, tmux, and curl to manipulate the local environment, monitor processes, and interact with network services.
[PRIVILEGE_ESCALATION]: The ao gc prepare command is designed to bypass interactive security dialogs in the Codex application. It achieves this by programmatically modifying the $CODEX_HOME/config.toml file to mark specific directories and hooks as trusted, allowing automation to proceed without user confirmation.
[EXTERNAL_DOWNLOADS]: The instructions direct the user to add and use an external "community" registry (https://registry.gascity.com/registry.toml) for downloading software packs. This source is not verified as a trusted vendor, posing a risk of downloading unvetted content.
[DYNAMIC_EXECUTION]: The prepare command generates and installs "AgentOps-owned wrappers" at specific formula check paths within the rig's .gc runtime. This involves creating executable files locally to intercept or extend runtime behavior.
[INDIRECT_PROMPT_INJECTION]: The skill implements a monitoring loop that reads data from untrusted sources, such as the gc mail inbox and tmux console panes. This data is used to drive agent actions without explicit sanitization.
Ingestion points: gc mail inbox, tmux capture-pane, and supervisor API responses.
Boundary markers: None identified; instructions suggest acting on signals found in prose.
Capability inventory: Significant operational capabilities including operate_gas_city and configure_codex_trust.
Sanitization: No evidence of data validation or instruction filtering for ingested content.

using-gc