- Home
- Skills
- Agents & Orchestration
- plan-out
Plan Out
A Claude skill that plans big features as small, non-contradicting spec modules with a decision log, and cleans up docs that have drifted out of sync.
$20
Works with the AI Agents you already use
plan-out
Example session with this skill installed
Let's plan the new user notification system. We need to support email and push. Specify the modules and the first few decisions.
- Read your context and instructions
- Compiled the plan-out
I have initialized the planning structure for the Notification System. I've created the core modules and logged the initial architectural decisions regarding transport types and provider selection.
Notification System Plan
Decisions Log (docs/decisions.md)
| ID | Decision | Owner | Status |
|---|---|---|---|
| D001 | Support Email and Push as primary transports | modules/transports | active |
| D002 | Use a unified provider interface for notification delivery | modules/engine | active |
Proposed Modules
docs/modules/engine.md: Logic for routing notifications based on user preferences.docs/modules/transports.md: Schema for email and push payloads.docs/modules/preferences.md: User-level opt-in/out state management.
Initial Task List (docs/tasks/notification-mvp.md)
# Task: Notification MVP
- [ ] Define notification payload schema in `modules/transports` [implements: transports#schema]
- [ ] Create abstract provider interface in `modules/engine` [implements: engine#interface]
- [ ] Log default retry policy in `decisions.md`
Consistency Check Result
python scripts/check.py check --docs docs- Result: PASS (3 modules created, 2 decisions active, 0 contradictions).
Next steps
- Review the
transportsmodule to confirm payload fields for Push notifications. - Decide on the specific third-party provider for Email (SendGrid vs SES) so I can log D003.
- Approve the task list to begin implementation.
Connects securely to your tools. The creator never sees your data.
About this skill
plan-out
A skill for Claude that plans big features as small, scoped spec modules, keeps them consistent as plans change, and cleans up docs folders that have drifted into contradiction.
The problem
Plan a feature over several AI sessions and you end up with a pile of files that are each internally consistent but disagree with each other. Session 3 changes a decision from session 1, nothing updates session 1's file, and both claim to be true. Asking an AI to reconcile them makes it worse: it can't tell the current decision from a superseded one, so it guesses, blends, and invents.
How plan-out prevents it
- One owner per fact. Every fact lives in exactly one module; everything else links to it.
- A decision log with status. Each decision has an id, an owner, and a status of
activeorsuperseded. Changes supersede, never overwrite. - Specs and tasks stay separate. Modules are the lasting documentation. Tasks are derived from them, link back one-way, and are archived when done.
- A script is the referee.
check.pymechanically catches broken links, stale decision references, duplicate definitions, uncovered sections and more, so consistency does not depend on an AI's impression. - Never assume. Unknowns become batched questions with recommended answers. Skipped questions are logged as explicit defaults.
- Cleanup is extractive. Existing docs are moved verbatim into the new structure, conflicts are decided by you, and a ledger proves nothing was lost or invented.
What it does
| You say | It does |
|---|---|
| "Plan a feature that lets users upload files" (vague or detailed) | Plays back its understanding, asks clarifying questions in small batches, proposes modules for approval, then writes modules, a decision log and a task list |
| "Change the upload limit to 25 MB" | Finds the owning module, shows the impact, supersedes the old decision, edits only what must change, flags affected tasks |
| "Our docs contradict each other, clean them up" | Inventories, extracts claims, shows you each conflict, migrates verbatim on a git branch, verifies with a ledger, then deletes the old files |
The result is docs a new engineer can learn the codebase from, plus a task list an AI agent can implement without asking anyone anything.
Output layout
docs/
README.md generated index
modules/ one file per module (the lasting spec)
decisions.md decision log
tasks/ active task lists
archive/ finished work, frozen
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.
Requires Python to run the check script. Documentation must follow specific templates in the docs/ folder to maintain consistency.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
Trust & safety
Security scan
Security scan passed 18 days ago
- Software
Python
Git