From great-minds
Structured phase planning with XML task plans. Researches the codebase, generates atomic task plans in XML format, verifies them against requirements, and organizes them into dependency-aware execution waves. Inspired by GSD (Get Shit Done) methodology.
How this skill is triggered — by the user, by Claude, or both
Slash command
/great-minds:agency-plan <phase-number or requirements-path><phase-number or requirements-path>This skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Plan a phase of work using structured XML task plans verified against requirements.
Plan a phase of work using structured XML task plans verified against requirements.
This skill brings GSD (Get Shit Done) concepts into the Great Minds agency:
Read $ARGUMENTS to determine:
If no phase specified, read STATUS.md to find the current/next phase.
Launch parallel research agents (model: haiku) to gather context:
Each agent reports back with structured findings. Do NOT proceed until research completes.
For each atomic requirement, create a task plan in XML format:
<task-plan id="phase-{N}-task-{M}" wave="{W}">
<title>{Short descriptive title}</title>
<requirement>{Traced requirement from spec}</requirement>
<description>{What this task accomplishes and why}</description>
<context>
<file path="{path}" reason="{why this file matters}" />
<!-- List ALL files the executor needs to read -->
</context>
<steps>
<step order="1">{Specific, atomic instruction}</step>
<step order="2">{Next instruction}</step>
<!-- Each step should be independently verifiable -->
</steps>
<verification>
<check type="build">npm run build</check>
<check type="test">npm run test -- --grep "{pattern}"</check>
<check type="manual">{What to visually confirm}</check>
</verification>
<dependencies>
<depends-on task-id="phase-{N}-task-{X}" reason="{why}" />
<!-- Empty if task is independent (wave 1) -->
</dependencies>
<commit-message>{Conventional commit message for this task}</commit-message>
</task-plan>
Group tasks by dependency:
Wave 1: [task-1, task-2, task-3] ← parallel
Wave 2: [task-4, task-5] ← parallel, after wave 1
Wave 3: [task-6] ← after wave 2
Launch a verification agent to cross-check:
Report any gaps back. Fix before proceeding.
Save to engineering/phase-{N}-plan.md:
# Phase {N} Plan — {Phase Title}
**Generated**: {date}
**Requirements**: {source file}
**Total Tasks**: {count}
**Waves**: {count}
## Requirements Traceability
| Requirement | Task(s) | Wave |
|-------------|---------|------|
| ... | ... | ... |
## Wave Execution Order
### Wave 1 (Parallel)
{XML task plans for wave 1}
### Wave 2 (Parallel, after Wave 1)
{XML task plans for wave 2}
...
## Risk Notes
{Any concerns from the risk scanner}
Also update STATUS.md:
phase: {N}
state: planned
plan: engineering/phase-{N}-plan.md
tasks_total: {count}
tasks_complete: 0
After the phase plan is written, spawn a haiku sub-agent as Sara Blakely:
Agent(model: "haiku", subagent_type: "sara-blakely-growth",
prompt: "Read engineering/phase-{N}-plan.md and the original PRD.
Gut-check from a real customer's perspective. Answer honestly:
- Would a real customer pay for this?
- What would make them say 'shut up and take my money'?
- What feels like engineering vanity vs. customer value?
Write to .planning/sara-blakely-review.md")
If Sara flags major customer-value gaps, the orchestrator should review the plan before proceeding to /agency-execute.
Provides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
npx claudepluginhub sethshoultes/great-minds-plugin --plugin great-minds