skills.rest
sharosoo
sharosooCOMMUNITY
Explore all skills by this author →
GitHub

ulw-plan

Generates decision-complete implementation plans from codebase exploration and user approval gates.

Updated Jul 31, 2024
Search Tags:#planning#subagents#software architecture#codebase exploration#work breakdown#pre-coding#approval gate
One-click install
npx skills add https://github.com/sharosoo/dotfile --skill ulw-plan-sharosoo
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: ulw-plan
Source: https://github.com/sharosoo/dotfile/tree/main/omp/omo-layer/skills/ulw-plan
Command: npx skills add https://github.com/sharosoo/dotfile --skill ulw-plan-sharosoo

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve? Large or vague coding requests often lead to premature implementation, missed requirements, and endless back-and-forth clarification. This Skill turns an ambiguous or multi-module request into one decision-complete work plan that a separate worker session can execute with zero further interview. ## Core Features & Use Cases - Explore-first planning: Grounds every decision in the actual codebase using read/grep/glob, optional codegraph MCP, and read-only research subagents before asking the user anything. - Intent routing: Classifies requests as CLEAR or UNCLEAR, asking only genuine owner-decision forks for clear intents and adopting announced best-practice defaults for fuzzy briefs. - Approval-gated plan generation: Scaffolds plan and draft files via a workspace-confined script, waits for explicit user approval, and supports optional high-accuracy dual review before handoff. - Use Case: You ask to "make auth better" across a large repo. The Skill explores the codebase, picks defensible defaults, writes a plan with per-task QA and acceptance criteria under plans/, and stops — a separate worker session executes it later. ## Quick Start Ask the agent to plan the refactoring of the authentication module before writing any code.

Frequently Asked Questions about ulw-plan

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

FAQPage Schema
How do I plan a large coding task before implementation?▼

Invoke the planning skill with your request and it explores the codebase first, asks only questions the code cannot answer, then writes a decision-complete plan under plans/ after your explicit approval. It never implements code itself.

What is the difference between CLEAR and UNCLEAR intent routing?▼

CLEAR intent means you know the desired outcome, so the skill asks only genuine owner-decision forks. UNCLEAR intent means the goal is fuzzy, so it researches maximally, adopts and announces best-practice defaults, and skips extra questions.

Does the planning skill write or modify product code?▼

No. The skill is strictly a planner: it only reads the codebase, runs read-only research subagents, and writes plan artifacts under plans/ and drafts/. Implementation is delegated to a separate worker session started by the user.

Can I resume a planning session after context compaction?▼

Yes. The skill records intent, decisions, approval state, and ledgers in a durable draft file, and the scaffold script is a safe no-op on re-run, so resuming reads the draft instead of restarting exploration.

When does the high-accuracy plan review trigger?▼

It triggers when you request high accuracy or deep review in any turn, or automatically for UNCLEAR intents. The review runs a dual check with momus and an independent oracle before the plan is handed off.

Related Skills

Explore More Software Engineering Skills→
aziglerazigler
8

test

Generate executable unit or integration tests from spec bead Section 5 cases.

Community
Advanced
v36372v36372

write-todos

Generate self-contained todos with constraints and acceptance criteria from plan artifacts.

Community
Intermediate
uzumaki-incuzumaki-inc

plan-eng-review

Review architecture, data flow, and testing strategy to lock in an engineering execution plan.

Official
Advanced
HonigbartHonigbart
1

quick-plan

Write executable plans to .ai/plans/<slug>/final_plan.md.

Community
Basic