Creates structured plans from requirements. Generates comprehensive plans with steps, dependencies, risks, and success criteria. Coordinates with specialist agents for planning input and validates plan completeness. Uses template-renderer for formatted output.
Instrucciones de origen · Vista previa de solo lectura
name
plan-generator
description
Creates structured plans from requirements. Generates comprehensive plans with steps, dependencies, risks, and success criteria. Coordinates with specialist agents for planning input and validates plan completeness. Uses template-renderer for formatted output.
["Coordinate with Analyst, PM, Architect for planning input","Break down requirements into actionable steps (<=7 per section)","Identify dependencies and sequencing","Assess risks with mitigation strategies","Validate plan completeness and feasibility","Use template-renderer for formatted plan output"]
error_handling
graceful
streaming
supported
related_skills
["template-renderer","writing-plans"]
verified
false
lastVerifiedAt
"2026-02-19T05:29:09.098Z"
source
builtin
trust_score
100
provenance_sha
7455ae00951a0074
Plan Generator Skill
Plan Generator Skill - Creates structured, validated plans from requirements by coordinating with specialist agents and generating comprehensive planning artifacts.
- Creating plans for new features
- Planning refactoring efforts
- Planning system migrations
- Planning architecture changes
- Breaking down complex requirements
- Validating existing plans
Every plan emitted by plan-generator MUST contain exactly these 6 top-level sections in this exact order:
## Problem — why this work matters; the pain point or gap being addressed
## Decision — what is being done; the chosen approach (not the steps, just the choice)
## Scope — what is IN scope; explicitly list what is OUT of scope
## Risks — what could go wrong; include mitigations and rollback commands
## Steps — ordered implementation checklist using - [ ] task format
## Done Criteria — acceptance criteria; objective conditions that prove the work is complete
Rules:
All 6 sections MUST be present — plans missing any section are invalid
Sections MUST appear in the order listed above — reordering is invalid
Each section must have substantive content (not a placeholder like "TBD")
Sub-sections and additional detail are allowed within each section
The pre-completion-validation hook enforces this order at task completion (severity: warning)
Why fixed order? AI-generated plans frequently reorder sections or omit scope/risks, causing incomplete execution. Fixed ordering enables automated validation and catches AI-slopped plans before they reach execution.
Iron Laws
ALWAYS include an executable command in every task — a task without a command is a wish, not an action; plans with wishes stall at execution because agents have nothing to run.
NEVER create a phase with more than 7 tasks — phases exceeding 7 tasks cause cognitive overload, lose parallel opportunities, and make rollback scoping impossible.
ALWAYS include a verification gate between every phase — proceeding without verification means failures in Phase N are discovered in Phase N+3, making rollback exponentially more complex.
NEVER generate a plan without explicit rollback commands for destructive operations — a plan without rollback paths is unrecoverable; rollback must be specified before the operation is attempted.
ALWAYS coordinate specialist input (Analyst, PM, Architect) before finalizing a plan — plans generated without domain expert review routinely miss implicit requirements, producing incomplete execution paths.
ALWAYS include a mandatory "Final: Self-review + token report" task as the LAST task in every plan — without this task, pipelines complete without cost visibility or quality self-assessment, and post-pipeline hooks never trigger.
ALWAYS emit the 6 canonical sections (Problem → Decision → Scope → Risks → Steps → Done Criteria) in exactly that order — plans with missing or reordered sections are rejected by pre-completion-validation.
MANDATORY FINAL TASK (IRON LAW — NEVER SKIP)
Every plan MUST include this as the LAST task, blocked by all other tasks:
Task: "Final: Self-review + token report"
Description: "Run ccusage to display today's token/cost (npx ccusage@latest --model --today). Then self-review: Can I improve any deliverable from this pipeline? If yes, create a reflection via reflection-spawn-request.json. NEVER dismiss failures as pre-existing. Mark metadata.isFinalTask = true and metadata.pipelineComplete = true in TaskUpdate(completed)."
This task is NEVER optional. It is part of every plan regardless of complexity.
The task must set metadata.isFinalTask = true so post-pipeline hooks trigger.
Example TaskCreate for planners to emit:
TaskCreate({
subject: "Final: Self-review + token report",
description: "1. Run: npx ccusage@latest --model --today — display token costs\n2. Self-review all deliverables: Can I improve this?\n3. If improvements found, queue reflection-spawn-request\n4. NEVER dismiss failures as pre-existing\n5. TaskUpdate(completed, { metadata: { isFinalTask: true, pipelineComplete: true } })",
activeForm: "Running self-review and token report"
})
Rules
The Iron Law of Planning
EVERY TASK MUST HAVE AN EXECUTABLE COMMAND
A task without a command is not a task - it's a wish.
Mandatory Elements
Every task must have: checkbox, ID, time estimate, command, verify
Every phase must have: verification gate, error handling
Every risk must have: rollback command
Anti-Patterns (DO NOT)
Anti-Pattern
Problem
Fix
"Install X" without command
Not executable
Add: cp -r source dest
"Verify Y works"
Vague
Add: npm test | grep PASS
"Update Z"
What file? What change?
Add exact Edit or sed command
No time estimates
Can't track progress
Add (~X min) to every task
No rollback
Can't recover from failure
Add rollback command
Quality Checklist
Before finalizing any plan, verify:
Can I copy-paste every command and run it?
Does every verify command have a clear pass/fail output?
Does each task have a verify command that proves completion objectively?
Does each task have a done criteria that is measurable and unambiguous?
Is there a rollback for every destructive operation?
Are time estimates realistic and granular?
Are parallel tasks marked with ⚡?
Template Integration
This skill uses the template-renderer skill to generate formatted plans:
Integration Flow:
plan-generator creates structured plan data (JSON)
