plan

📁 superinterface-labs/superpaper 📅 Today
2
总安装量
1
周安装量
#66698
全站排名
安装命令
npx skills add https://github.com/superinterface-labs/superpaper --skill plan

Agent 安装分布

amp 1
cline 1
opencode 1
cursor 1
kimi-cli 1
codex 1

Skill 文档

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).