skill-executing-plans

Executes approved implementation plans with Git reconciliation and per-task verification.

Updated Apr 9, 2026
One-click install
npx skills add https://github.com/longdang193/project-OS-starter --skill skill-executing-plans-longdang193
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: skill-executing-plans
Source: https://github.com/longdang193/project-OS-starter/tree/main/generated_agents/claude/skills/skill-executing-plans
Command: npx skills add https://github.com/longdang193/project-OS-starter --skill skill-executing-plans-longdang193

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Approved implementation plans often drift from repository reality during execution, causing scope creep, unverified completion claims, and lost work. This Skill enforces disciplined plan execution by reconciling plan state against Git evidence, verifying each task before advancing, and preserving unrelated user changes. ## Core Features & Use Cases - Plan-Git Reconciliation: Reads workflow state from the plan and validates it against current Git evidence, blocking on mismatch instead of inferring progress. - Bounded Task Execution: Executes one task at a time with focused verification, diff inspection for scope growth, and explicit divergence handling. - Executor Coordination: Resolves executor and template profile values for Git-tracked coordinated work across Codex, DeepAgents, and Tura runtimes. - Use Case: A lead controller agent resumes a partially completed multi-task plan, reconciles the task ledger against the current branch state, executes the next unblocked task with focused tests, and hands off to completion verification. ## Quick Start Execute the approved implementation plan in docs/plans/current-plan.md, verifying each task against repository evidence before marking it complete.

Frequently Asked Questions about skill-executing-plans

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I execute an approved implementation plan with an AI coding agent?▼

Read the full plan first, reconcile its workflow state against current Git evidence, then execute the smallest unblocked task with focused verification. Update task status only after repository proof supports the transition, and hand off to completion verification when all tasks finish.

What should I do when a plan no longer matches the repository?▼

Correct stale file paths or command names directly in the plan for small mismatches. For changed scope, unresolved design, or missing acceptance criteria, stop and return the plan for revision rather than inventing design during execution.

Can plan execution resume after a session interruption?▼

Yes, but continuation requires settled prior ownership, explicit controller authority, and reconciliation of the task ledger against Git evidence. Never recover progress from thread or session state alone; leave a compact handoff with changed files and next actions.

When should plan execution stop and request direction?▼

Stop when approved scope is ambiguous, a destructive action lacks authorization, user work conflicts with planned edits, credentials are unavailable, or repeated verification failures need a design decision. Do not rerun unchanged failing commands without a new hypothesis.

Does this approach support parallel or delegated task execution?▼

Yes, through companion skills for subagent-driven development and parallel lane dispatch when tasks have disjoint write ownership. Executor selection stays explicit via the task ledger, and delegation requires the runtime grant to allow child agents.