Execute is a routing skill that scans text files for inline <@skill-name: prompt> annotations and dispatches each tagged prompt to the corresponding skill subagent. It standardizes a subagent contract (file path, surrounding text, prompt) and defines a sequential processing workflow so subagents can apply changes directly to files.
Use Execute when you want to orchestrate modular operations inside documents: batch-edit files, run proofreads, produce figures, fill placeholders, or trigger other skill workflows from annotated text. It is designed for programmatic, stepwise processing where each tag may spawn an independent subagent run.
Inferrable: agents that implement subagent spawning and file I/O (Copilot/Codex-style runners, agent frameworks that support file-based workflows).
A router skill that scans files for <@skill: prompt> tags and dispatches each to a matching subagent. No scripts bundled — purely a SKILL.md that defines a dispatch workflow with a registry table and subagent contract. Well-structured with clear rules, but relies entirely on other skills being present to function.
Static SKILL.md only, no scripts. The dispatch pattern is clean but the registry is hardcoded in the markdown — adding new skills requires editing the file. The skill explicitly warns about false triggers from demonstration tags, which shows good defensive design.