plan

name	plan
description	Decompose a goal into a phased implementation strategy
type	skill
status	active
version	2
created	2026-02-07
tags	["type/skill"]

plan

Decompose a goal into a phased implementation strategy.

Part of [[.agents/skills/AGENTS|Skills Directory]].

When to Use

• New project or initiative needs a roadmap
• Complex task needs breakdown before execution
• Strategic decision requires options analysis

Inputs

Input	Required	Description
goal	yes	What needs to be accomplished
constraints	no	Budget, timeline, tech stack, team size
context	no	Links to relevant notes in entity folders

Process

1. Clarify the goal — what does success look like? State observable outcomes, not internal attributes.
2. Identify unknowns — what do we need to research or prototype first? For significant unknowns, plan explicit spike/prototype phases that validate feasibility before committing.
3. Break into phases with clear dependencies
4. For each phase: define deliverables and acceptance criteria phrased as behavior a human can verify
5. Identify risks and mitigations
6. Write the plan as a note or add tasks to the board
7. For complex tasks: write the plan as a file (in project folder or .plans/). Plans that live as files can be handed to any agent or human cold.

Subagent Orchestration

When work partitions into independent subgraphs, dispatch parallel subagents with clear scope boundaries:

• Each subagent gets a self-contained task spec: objective, owned areas, do-not-touch list, acceptance criteria
• Define shared interfaces/contracts before parallel work begins
• Integrate in controlled order (highest-dependency first)
• Do NOT parallelize tightly-coupled work through shared state

Outputs

• Plan document (in project folder or .plans/)
• Tasks on [[apps/My tasks]] with phase labels
• Dependencies noted (which tasks block which)

Conventions

• Plans are living documents — update as reality changes. Maintain a Decision Log for why you changed course.
• Prefer fewer, larger phases over many tiny steps
• Every phase should produce something demonstrable — observable outcomes, not just "code that compiles"
• Include a "What We Don't Build" section to prevent scope creep
• Record surprises and discoveries as you go — they inform future plans
• Plans should be self-contained enough that a novice (or stateless agent) can pick them up cold
• If a plan will produce many files in one folder, plan subfolder structure upfront. Reorganize any folder that crosses ~8–10 items (per [[AGENTS]] conventions).