Diversio Engineering
On this page
SKILL PROCESS

Plan Directory

Create or update a requested plan directory with PLAN.md and numbered task files.

Overview

Create or update a requested plan directory with PLAN.md and numbered task files.

This skill ships inside the Plan Directory plugin and can be installed through the Claude Code marketplace or directly in Codex from its skill path.

Parent Surface

Parent docs: Plan Directory

Related wrapper commands from the parent plugin:

/plan-directory:plan

When to Use This Skill

structured, step-by-step plan.

explicit, verifiable checkboxes.

or archiving completed plans.

planning discussion or an implementation task.

Do not use this skill for ad-hoc todo lists, single-file notes, or when the user explicitly wants a different format.

  • Scaffolding a new project, feature, or migration that benefits from a structured, step-by-step plan.
  • Creating a repeatable plan that an LLM or engineer can execute with explicit, verifiable checkboxes.
  • Maintaining an existing plan directory: adding tasks, updating progress, or archiving completed plans.
  • When the user requests a structured plan artifact, not merely a short planning discussion or an implementation task.

LLM Intake (Required Inputs)

Derive these inputs from the request, existing plan, and repository. Choose a clear title and slug and propose the task breakdown yourself. Ask only when missing scope, conflicting decisions, or an ambiguous destination would change the plan; do not require the user to supply the plan you were asked to create.

When updating an existing plan, read the current PLAN.md first to understand context before modifying.

Invariants (Do Not Violate)

usage instructions, locked decisions, and a task index with checkboxes. Detailed steps live in individual task files.

is zero-padded (001, 002, ..., 999) and is hyphenated lowercase.

Completion Criteria, and Dependencies. All are required (use "None" for Dependencies if truly independent, "N/A" for Tests only with justification).

fields, emoji indicators, or separate progress sections. Check - [x] when complete, leave - [ ] when pending.

are checked, check the corresponding task in PLAN.md.

work has started, it is permanent. Append new tasks at the end.

from the file alone without reading other task files.

section describing what's blocking and link to the blocking task or issue.

Workflow

state, and make targeted modifications.

merely its task breakdown, remains unclear.

For new plans:

For updates:

Keep it minimal. Include only:

Each task file must include:

Before delivering a new or updated plan, read it again as if you did not write it and fix obvious issues immediately. Look for:

Do not wait for the user to ask for this pass. It is part of the skill.

As work completes:

  • Create: No plan directory exists. Scaffold everything from scratch.
  • Update: PLAN.md already exists. Read it, understand the current state, and make targeted modifications.
  • Confirm all required inputs are present.
  • Turn a clear goal into concrete tasks. Ask when the intended outcome, not merely its task breakdown, remains unclear.
  • For updates, verify the plan slug and location match the existing plan.
  • Purpose: 1-3 bullets on why the plan exists.
  • How to use: Brief instructions for working the plan.
  • Decisions (locked): Key constraints that should not change.
  • Task Index: Checkbox list linking to task files.
  • Completion: Definition of when the entire plan is done.
  • Goal: Single sentence describing the outcome.
  • Dependencies: List of task numbers that must complete first, or "None".
  • Scope: Bulleted list of what's in scope (and optionally out of scope).
  • Checklist: 3-8 concrete, actionable steps with checkboxes.
  • Tests: Specific test commands, files, or manual QA steps.
  • Completion Criteria: Measurable definitions of done.
  • Notes (optional): Constraints, references, warnings.
  • Blockers (optional): Added when work is blocked; removed when unblocked.
  • Missing or incorrect dependencies / execution order
  • Checklist items that are vague, oversized, or not verifiable
  • Tests that are missing, generic, or inconsistent with the task scope
  • Hidden blockers, assumptions, or locked decisions that were left implicit
  • Scope leaks where a task bundles unrelated work

Resources

Declared allowed tools:

ReadWriteEditGlobBash

References

  • advanced.md
  • templates.md

Installation

Switch between Claude Code and Codex, then copy the install command for the runtime you use.

claude plugin marketplace add DiversioTeam/agent-skills-marketplace
claude plugin install plan-directory@diversiotech

Invocation:

/plan-directory:plan