The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Node.js spawnSync to execute local binaries including codex, tmux, and custom local commands specified via the --resume-command flag in scripts/schedule.mjs.
This functionality is used to deliver scheduled prompts to the user's terminal or tmux session.
The implementation includes sanitizeCodexArgs, which uses a strict default-deny allowlist to permit only a small set of safe flags (like --model) while blocking flags that could weaken security sandboxes or approval policies.
[PROMPT_INJECTION]: The skill functions as a surface for indirect prompt injection because it ingests, stores, and later executes prompts provided by the user in scripts/schedule.mjs.
Ingestion points: Prompts are ingested through the add and parse-schedule CLI subcommands and stored in the local .codex/schedule/tasks.json file.
Boundary markers: The skill does not add specific delimiters or safety instructions around the prompt content before passing it to the runners, relying on the underlying tool's standard input handling.
Capability inventory: The script possesses the capability to execute subprocesses, write to the filesystem, and manage local process locks.
Sanitization: All arguments passed to subprocesses are validated against a strict character allowlist (regex: /^[-A-Za-z0-9_.,:=/@+]+$/) to prevent shell injection.

schedule