Skill

plan-generator

Generate a structured plan from a feature description. Spawns plan-generator agent to analyze the codebase and produce a task-by-task plan. Use when starting new work or to expand a brief description into a detailed plan. NOT for modifying existing plans.

Install

npx claudepluginhub hir4ta/qult --plugin qult

Tool Access

This skill uses the workspace's default tool permissions.

Preview

Generate a structured implementation plan from a brief feature description.

SKILL.md

Similar Skills

test-driven-development

1 file

Guides strict Test-Driven Development (TDD): write failing tests first for features, bugfixes, refactors before any production code. Enforces red-green-refactor cycle.

antigravity-bundle-qa-testing

32.8k

systematic-debugging

10 files

Guides systematic root cause investigation for bugs, test failures, unexpected behavior, performance issues, and build failures before proposing fixes.

antigravity-bundle-qa-testing

32.8k

ab-test-setup

Guides A/B test setup with mandatory gates for hypothesis validation, metrics definition, sample size calculation, and execution readiness checks.

antigravity-bundle-qa-testing

32.8k

Stats

Parent Repo Stars0

Parent Repo Forks0

Last CommitApr 5, 2026

Actions

View Source View Plugin View on GitHub View README

/qult:plan-generator

Generate a structured implementation plan from a brief feature description.

Stage 0: Spec document check

Before generating a plan, check if a spec document exists:

Search for docs/spec-*.md files using Glob
If found: read the most recent spec and include it as context for the plan-generator agent
If not found: continue without a spec, but note in the agent prompt that no spec was provided

This ensures that design decisions from /qult:explore are carried into the plan.

Stage 1: Plan generation (independent agent)

Detect the language of $ARGUMENTS (e.g., Japanese, English, Chinese, Korean, etc.). If the language is not English, include an explicit instruction in the agent prompt: Output language: <detected language name> (e.g., Output language: 日本語).

Spawn one plan-generator agent with:

The user's feature description: $ARGUMENTS
The spec document content (if found in Stage 0)
The output language instruction (if non-English detected)

The agent analyzes the codebase independently and outputs a complete plan in markdown format.

Stage 2: Persist

Write the final plan to .claude/plans/plan-<timestamp>.md.

Use format: plan-YYYYMMDD-HHMMSS.md for the filename.

Stage 3: Plan evaluation (independent agent)

Spawn one plan-evaluator agent with the plan file path from Stage 2.

The evaluator reads the plan file and scores it on three dimensions:

Feasibility: Can Claude Code execute each task as described?
Completeness: Are all affected files covered, including consumers and tests?
Clarity: Is each task unambiguous and actionable?

If the evaluator's verdict is Plan: REVISE:

Read the evaluator's findings
Fix the issues in the plan file
Re-spawn the evaluator on the updated plan

The SubagentStop hook enforces the score threshold mechanically (aggregate >= 10/15, max 2 iterations).

Output

Summary line: Plan generated: .claude/plans/<filename> (N tasks, evaluation: Feasibility=N Completeness=N Clarity=N)

Then suggest: "Enter plan mode (Shift+Tab x2) to review and approve the plan."