Sigma Playbook

Sigma · Prompts

Feature Run

Run a feature end-to-end across tiered Claude models — planning at the top tier, implementation in the middle, deterministic verification and read-heavy recon at the bottom — without burning the main session's context or overpaying for mechanical work. The entry point to the sigma-feature-run skill.

Usage: /sigma-engineering:feature-run [mode] <feature name or plan path> Modes: plan · execute · verify · review · gather · report. Omit the mode and one is inferred from repo state.

Install and run

It ships with the Sigma engineering plugin. Once the plugin is installed:

  • The /sigma-engineering:feature-run slash command drives a run — source: commands/feature-run.md.
  • The sigma-feature-run skill auto-activates whenever a task spans planning, implementation and verification — even if you never type the command.

What it does

Routes on intent across six modes, delegates read-heavy and deterministic work to subagents, and keeps decisions — not transcripts — in the main session. The rule that matters most: capability flows down, evidence flows up; nothing below the author's tier signs off on the author's design, and a shell command beats a model call whenever one answers the question.