Skip to content
Back to open source

Deep Work Plan

Flagship

The gated-plan methodology any AI coding agent can execute.

Deep Work Plan is a full agent methodology packaged as an open-source skill pack. Give your AI coding agent an ambiguous multi-step goal — "rebuild the developer portal", "migrate the payments service", "add SAML" — and it produces a numbered, atomic, validated plan under `.dwp/plans/`. Every task carries an explicit validation gate. The agent runs one task at a time and refuses to move on until validation passes. No lost progress. No half-done tasks masquerading as done.

License: MITPrimary language: Bash / Markdown

8

sub-skills for the full lifecycle

5+

AI coding agents supported

0

silent failures — every gate is explicit

MIT

licensed, end to end

Why it exists

The bits that make it worth adopting.

Numbered atomic tasks

Each task ships one deliverable. No 'build the whole feature' catch-alls. Every step is small enough that an agent can execute it deterministically and a human can review it in a single pass.

Explicit validation gates

Every task carries a criterion for 'done' — a passing test, a green build, a specific file existing, a curl returning 200. The agent refuses to advance until the gate passes. 'Done' means actually done.

State that survives sessions

Plan state lives in `.dwp/plans/PLAN_{name}/state.json`. If the agent crashes, the laptop reboots, or you swap Cursor for Claude Code, the plan resumes on the exact task it stopped on.

Refinable without losing work

Add tasks in the middle, remove ones that turned out irrelevant, reorder for a better dependency graph — `refine` preserves everything already completed.

Cross-agent by design

Same skill install, same plan structure, same gates across Claude Code, Cursor, Codex, Gemini, Copilot. Swap agents mid-plan and the methodology doesn't change.

Optional Dailybot addon

Wire plan kickoff, mid-plan progress, blockers, and completion into your team's standup. Kickoff, blocker, and completion reports are first-class Dailybot updates. Disable the addon if you don't use Dailybot — the methodology stands on its own.

Install

Get started in seconds.

Pick the channel that matches your stack. Every path lands you in the same working state.

npx skills add DailybotHQ/deepworkplan-skill

In practice

What teams actually do with it.

01

Rebuild a developer portal

One fuzzy goal, one numbered plan. Twelve atomic tasks, each with its gate. The agent ships one PR-sized unit at a time — reviewable, revertable, resumable. No 'we shipped 40% but we're not sure which 40%'.

02

Migrate a legacy service

The kind of work that used to sprawl for weeks with no clear 'am I 30% or 80% done?' answer. DWP produces the plan, and the human always knows exactly which tasks are complete, which are blocked, and which are the critical path.

03

Refactor across many files

Twenty-plus files touched, each with a small change. Numbered tasks and validation gates make a big refactor feel like small refactors done in sequence — no half-migrated state, no orphaned commits.

04

Onboard a new agent to an existing repo

`dwp-onboard` runs the same setup a human would do — writes AGENTS.md sections, adds skills, wires hooks — and reconciles with what's already there. Safe to run on a repo that already has DWP configured.

At a glance

The short list.

Eight composable sub-skills

Works with Claude Code, Cursor, Codex, Gemini, Copilot

Optional Dailybot progress-report addon

MIT-licensed

FAQ

The questions we hear the most.

Do I need Dailybot to use DWP?

No. DWP is a standalone methodology. The Dailybot addon is optional — if you use Dailybot, plan lifecycle events show up in your standup automatically. If you don't, the plan still runs, the gates still enforce, the state still survives.

Where do plans live on disk?

`.dwp/plans/PLAN_{name}/` — a directory per plan with tasks, state, and progress. It's a normal folder of Markdown and JSON, so any editor or grep works on it.

What if a task fails its gate?

The agent stops, reports the failure, and refuses to move on. Fixing the failure is a real, visible task — not a silent skip. This is what makes 'done' actually mean done.

What it is

Deep Work Plan is an open-source agent methodology packaged as a skill pack. It teaches your AI coding agent how to ship multi-step work without losing the plot — same rigor a senior human engineer would apply to a large refactor or migration, translated into a format an agent can execute deterministically.

Why open source

Because a methodology only matters if people can inspect, criticize, and adapt it. DWP is MIT-licensed and developed in the open. There’s a companion site at deepworkplan.com that documents the full methodology; the skill is the executable expression of what that site describes.

How it fits the ecosystem

DWP is standalone — it works with any AI coding agent, whether or not you use Dailybot. If you do use Dailybot, the optional addon reports plan lifecycle events (kickoff, blocker, completion) via the Agent Skill and the dailybot CLI, so your team follows big multi-week work in the same standup feed they already read.

Ready to try it?

Open source, MIT-licensed, and shipping in production at Dailybot every day. Fork it, wire it in, contribute back.