
from ystack8
Execute a previously created PLAN.md: run tasks in order (waves), produce atomic commits, and verify each step — designed for doc‑driven code execution workflow
This skill implements the execution phase for a doc‑driven development harness. Given a PLAN.md produced by a prior build step, the skill runs tasks in dependency waves, executes each task (often via fresh subagents), verifies results (typecheck, tests), and creates atomic commits with progress updates.
Use when the user says 'go', '/go', 'execute the plan', or otherwise confirms they want the agent to implement a prepared PLAN.md. Intended for automated code execution workflows where plans are explicit.
Designed for code‑execution focused agents such as Claude Code or other runtimes that support subagents or programmatic execution.
ystack-go is a plan execution skill for the ystack agent harness. It reads a PLAN.md, computes execution order with dependency waves, runs tasks sequentially, and produces atomic git commits per task. No bundled scripts — purely instruction-driven. The SKILL.md is well-structured with clear phases, deviation handling rules, and verification steps. Niche but practical for ystack users.
Clean skill with no security concerns. No scripts to audit. Well-documented execution workflow with sensible deviation handling. Tightly coupled to ystack ecosystem.