Delegating to Agents
Adapted by Lex from SigmaStudioz from davidondrej/skills · MIT · version 1.0.1.
Hand bounded work to another AI agent CLI or TUI (Codex, Cursor, Pi, Claude Code, Hermes) and
keep control of it: a complete brief, an isolated git worktree when others have uncommitted
work, progress lines you can watch as events, BLOCKED questions forwarded to whoever owns the
decision, and results reviewed before they are reported back up the chain.
What a brief must contain comes from the bundled handoff contract; this skill covers the mechanics of driving each agent.
Changes in 1.0.1
Bundles the handoff contract inside the installed skill and resolves every operative guide link locally. The versioned adaptation preserves authorization, briefs, isolation, evidence and reporting rules without a mutable remote instruction dependency. No helper or service added.
Changes in 1.0.0
First release in this collection. Adds Cursor agent CLI usage, sandboxed modes instead of run-everything flags, worktree isolation, a progress-file and monitor pattern, and reporting back to the delegating agent. Personal account details from the earlier local copy were removed.
Install
npx skills add sigmastudioz/agent-skills --skill delegating-to-agents
Guidance only: no executable helper or required service. Driving another agent runs commands on your machine; the skill asks for approval before commands, remote access or file changes.