The Agent Skills Directory

[COMMAND_EXECUTION]: Generates and executes temporary shell scripts to facilitate TUI testing under tmux.\n
Evidence: scripts/tui-smoke.sh creates a temporary tui-launch.sh script to boot the application in a headless session.\n- [COMMAND_EXECUTION]: Spawns local binaries and background services using dynamic arguments and environment overrides.\n
Evidence: scripts/lib/app-server-client.mjs uses the Node.js spawn method to drive the codex binary, and scripts/lib/common.sh manages the local mock model process.\n- [EXTERNAL_DOWNLOADS]: References official software packages and standard utilities for environment configuration.\n
Evidence: references/docker-qa.md lists dependencies including @openai/codex and opencode-ai from official package registries.\n- [DATA_EXFILTRATION]: Accesses local configuration files to calculate integrity hashes for isolation verification.\n
Evidence: scripts/lib/common.sh reads ~/.codex/config.toml to ensure the user's primary configuration is not modified during testing.\n- [PROMPT_INJECTION]: Processes user-provided prompt data that is interpolated into the testing harness communication channel.\n
Ingestion points: scripts/lib/app-server-client.mjs (via the PROMPT environment variable).\n
Boundary markers: Absent.\n
Capability inventory: Controls the codex application via its programmatic app-server interface.\n
Sanitization: Employs JSON.stringify to escape inputs within the protocol messages, preventing command injection at the application protocol level.

codex-qa