Skip to content
Agentmart
Submit a tool

planning

Plan complex or creative work as a directed acyclic graph (DAG / mind-map) before executing, keeping the human as the decider at every gate.

Skill by scott-fryxell

IndexedMIT★ 125
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/scott-fryxell-planning

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Plan complex or creative work as a directed acyclic graph (DAG / mind-map) before executing, keeping the human as the decider at every gate. Use when a task branches, explodes like a mind map, or is a creative project - anything with parallel or dependent pieces a linear checklist would flatten. Produces an ordered, reviewable plan, checks in at each gate, then executes. Built from the Building an Advanced Agentic Harness article's Planner role, inverted so the harness plans and the human decides.

Use cases: writing, productivity

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/planning

What's inside

Skill name
planning
Description
Plan complex or creative work as a directed acyclic graph (DAG / mind-map) before executing, keeping the human as the decider at every gate. Use when a task branches, explodes like a mind map, or is a creative project - anything with parallel or dependent pieces a linear checklist would flatten. Produces an ordered, reviewable plan, checks in at each gate, then executes. Built from the Building an Advanced Agentic Harness article's Planner role, inverted so the harness plans and the human decides.
License
MIT
Metadata
category: Workflow
Instructions
181 lines in SKILL.md · view on GitHub

Files (1)

  • SKILL.md

SKILL.md (first lines)

# Planning

Plan work as a directed acyclic graph before doing it. The agent proposes, you
approve, then execution runs. This skill never runs a crew solo.

## Roles

Five fixed names, used across this skill, AGENTS.md, and plan files:

1. **Explorer** - figure out what the task really is. Skip once the ask is clear.
2. **Planner** - shape branchy work into a DAG and stop at Gate 1. Trivial or
   linear work passes straight through.
3. **Worker** - run nodes in topological order.
4. **Critic** - verify each node, cheapest deterministic check first.
5. **Promoter** - decide if verified work is worth showing off; most is not.

No aliases like "research phase" or "promote step."

## When to use

Work that branches, rejoins, or has dependencies between pieces. Skip trivial or
linear tasks - planning has a cost.

## Core rule

Nodes are units of work, edges are dependencies. Build the graph, check in at
each gate, and never self-run the whole DAG.

## Process

### 1. Planner - build the DAG

- One node per unit of work: id, one-line goal, effort (cheap/medium/expensive).
- Edges only where one node depends on another. No cycles - a cycle means the
  plan is muddled.
- Split anything that would become a paragraph to describe.

### 2. Planner - render the DAG

**Use Mermaid**, so it renders in Obsidian, GitHub, and editors:

````
```mermaid
flowchart LR
  a["a research angle"] --> b["b pick sources"]
  b --> d["d assemble"]
  c["c draft section 1"] --> d
  d --> G{{"🚦 GATE 1<br/>shape approved"}}

  classDef gate fill:#3a2f1a,stroke:#c9a227,color:#f5e6c8;
  classDef done fill:none,stroke:#555,stroke-dasharray:3 3,color:#777;
  classDef active fill:#1f2f3a,stroke:#4f8fbf,color:#d6e9f5;
  classDef blocked fill:#3a1f1f,stroke:#bf4f4f,color:#f5d6d6;
  class G gate;
  class a done;
  class b active;
```
````

Node ids lead the label (`a1`, `b3`) so the table can refer to them. Gates use
the `{{...}}` hexagon and the `gate` classDef. Keep labels to a few words and put
effort in the table, not the graph.

Before showing a DAG in chat, run `bin/mermaid-fit.js <file>`. Pi's feed shows
the raw fence when a diagram is wider than the pane; shorten labels until it
passes.

**Status is color plus emoji**, so it reads in any renderer and in plain text:

| Status      | Emoji | classDef                |
| ----------- | ----- | ----------------------- |
| done        | ✅    | `done` (dimmed, dashed) |
| in progress | 🔄    | `active` (blue)         |
| blocked     | ⛔    | `blocked` (red)         |
| not started | none  | default                 |
| gate        | 🚦    | `gate` (amber)          |

The emoji leads the label (`"✅ a1 research angle"`) and matches the node's
`class` line. Update both as work moves. Finished work dims so the eye lands on
what is left.

**The node list dims too.** A finished row's id is struck through,
`| ~~a1~~ ✅ |`. Markdown renders that as `<del>`, so any viewer dims the
whole row with `tr:has(del) { opacity: 0.45 }` and GitHub shows it crossed
out. One line above the table names the legend.

ASCII only for a throwaway sketch that will never persist.

**Put the DAG in a plan file.** `plans/<plan>.md`, or a companion
`plans/<plan>.plan.md` if that file should stay pure spec. One file per plan.
Gate decisions land in Decisions as one-liners, nowhere else.

### 3. Planner - keep it readable

Word bloat is the nemesis. Judge every line: **does this change a decision, or
could the code say it?** If neither, cut it. A plan over ~120 lines has absorbed
something - find it and delete it.

Write ideas, not explanations: one idea per line, seven to ten words where
possible. The reader knows what to do; no instructions addressed to them, no
hedges, no reasoning they did not ask for.

Open with who it is for: one sentence naming the target customer, so every
node can be judged against them.

Carries: decisions and what was rejected, one line each. Open questions, as
questions. What to do next and how to judge it. What the code cannot say -
gotchas that bite, why the thing exists at all.

Does not carry:

- A changelog. Git holds history. No dated "what changed" list, no gate log, no
  "approved on" section, no restating how the plan got here.
- Settled questions. Move the decision to Decisions and delete the question.
- Nodes a decision absorbed. When a branch is picked, the choosing node
  disappears and the choice stands alone.

Finished nodes stay, dimmed, until the human says to prune. Pruning removes
the node, its edges, its row, and any mention elsewhere. Never prune on your

From scott-fryxell/brayness, MIT.

More from scott-fryxell/brayness

Signals

Stars
125
Forks
15
Open issues
0
Last commit (this folder)
Sep 28, 2026 (29 days ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from scott-fryxell

Cite this page

planning skill. Agent Mart. https://agentmart.dev/skills/scott-fryxell-planning (last updated 2026-10-24)