DeepDone Skills
System nominal

Requirements to shipped code: automated.

DeepDone Skills

An automated control system for AI-assisted software work. Works best with Codex.

Implement end to end

Choose depth

One supervisor. Two useful operating modes.

full loop

End to end

Requirements go through intake, decisions, build, verification, review, safe fixups, and local commit when gates pass.

[$DeepDone Orchestrate] Build this end to end

bounded

Single step

One controlled transition only. Useful when you want intake, one milestone, verification, review, or a commit candidate.

[$DeepDone Orchestrate] Continue one step

Workflow

Plan, build, verify, release.

01

Plan

Shape work into a small task, one epic, or roadmap. Resume stale state before coding.

plansyncadvancedecide
02

Build

Execute one clear milestone. Scope stays narrow. Ledger tracks changed truth.

implement
03

Verify

Run smallest useful checks first. Review local diff for bugs, regressions, security, and missing tests.

verifyreviewfixup
04

Release

Create focused local commits, draft PR or MR material, inspect CI when asked, archive after merge or release.

commitprarchive

Skill rack

Compact units. Explicit outputs.

Supervisor

1

Orchestrate

deepdone-orchestrate

Default entry point for both end-to-end work and single-step control.

Next transition, completed run, local commit, or exact stop reason.

Use $deepdone-orchestrate. Mode: end-to-end.

State

4

Plan

deepdone-plan

Requirements are new, fuzzy, or large enough to need memory.

Small task, single epic ledger, or roadmap plus active ledger.

Use $deepdone-plan.

Sync

deepdone-sync

Thread resumed, diff exists, or next action may be stale.

True state, drift notes, corrected next action.

Use $deepdone-sync.

Advance

deepdone-advance

Current epic is complete and queued work remains.

Updated roadmap and exactly one new active ledger.

Use $deepdone-advance.

Decide

deepdone-decide

Architecture, library, API, deployment, or schema choice could be wrong from memory.

Decision block with chosen option, rejected options, sources, and verification impact.

Use $deepdone-decide.

Build

1

Implement

deepdone-implement

Next milestone is clear, acceptance is known, and touched files are obvious.

Focused code change, progress update, next verification path.

Use $deepdone-implement.

Gates

3

Verify

deepdone-verify

Code changed or a milestone claims done.

Pass, fail, or blocked verification log plus residual risk.

Use $deepdone-verify.

Review

deepdone-review

Implementation and verification are complete enough to challenge.

Findings by severity or review-result pass marker.

Use $deepdone-review.

Fixup

deepdone-fixup

Review found clear low-risk defects inside current milestone.

Focused fix, verification result, remaining blocker if any.

Use $deepdone-fixup.

Lifecycle

3

Commit

deepdone-commit

Work is ready to become one VCS unit.

Commit candidate or local commit when authorized.

Use $deepdone-commit.

PR

deepdone-pr

A local commit exists and work needs external review handoff.

Title, body, blockers, or CI summary. No push by default.

Use $deepdone-pr.

Archive

deepdone-archive

Work is merged or released and you explicitly want closeout.

Archived ledger, roadmap pointer update, post-merge truth.

Use $deepdone-archive.

Install and run

Use the orchestrator first.

The skills repo contains agent manifests, examples, doctor checks, and workflow docs. Clone it once, then start with the orchestrator.

One active thread of work beats parallel plan sprawl.

Every gate records truth where the next run can find it.

Push, merge, deploy, and archive stay explicit.

Install
SKILLS_REPO="https://github.com/lakesoftai/deepdone-agent-skills"
git clone "$SKILLS_REPO" ~/.agents/skills/deepdone
Run
Use $deepdone-orchestrate.
Requirements: <goal>
Mode: end-to-end.
Control
Mode: one-step
Mode: until-review
Mode: until-commit-candidate
DeepDone Skills

GitHub repository

https://github.com/lakesoftai/deepdone-agent-skills
Back to top