Supporting skill

`tcx-automate`

Create or update a Codex app Scheduled Task for repeated research, monitoring, review, or status work. The saved task executes the actual work skill, not this authoring skill again.

Design the recurring task #

Specify the cadence, desired output, notification behavior, and stop conditions. The schedule should describe a useful recurring question rather than an unbounded instruction to act.

Example

$tcx-automate
Research NVDA each weekday. Report material changes and stop before any order action.

What runs later #

The saved runtime prompt invokes the real work skill, normally $tcx-workflow. Every scheduled run is submitted through the same root-turn path as an interactive request; it remains subject to its own current sandbox, hooks, policy, and service gates.

That means a scheduled task is not recursively routed back to $tcx-automate, and an Automation origin is not itself a reason to trust an action.

Keep authority explicit #

Report-only is the default effect level. Ordinary recurring research, monitoring, analysis, portfolio/status review, draft, and assisted-execution prompts omit authority markers. Scheduled research never auto-promotes results into a Wiki or Brain.

Only a deliberately delegated final-action task uses the canonical plain $tcx-order-allow invocation on the first meaningful line. A deliberately delegated workspace Build task uses plain $tcx-build; a Brain, community Wiki, or Strategy lifecycle task uses plain $tcx-brain, $tcx-wiki, or $tcx-strategy first in trading-research. Saved prompts stay path-independent rather than embedding Markdown skill links. Markers are never combined, the task must pin the requested action and target, and each scheduled turn requires a fresh current-turn grant decision.