sdd

Executes plan.md tasks via fresh subagents with independent review gates before each commit.

Updated Aug 31, 2026
One-click install
npx skills add https://github.com/keefhub/travel-expense --skill sdd-keefhub
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: sdd
Source: https://github.com/keefhub/travel-expense/tree/main/.claude/skills/sdd
Command: npx skills add https://github.com/keefhub/travel-expense --skill sdd-keefhub

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Executing a multi-task implementation plan by hand risks skipped steps, unreviewed code, and lost progress when a session dies mid-plan. This Skill drives a plan.md file task-by-task using fresh subagents for implementation, mandatory independent review gates (spec compliance and code quality) before every commit, and file-based state (PROGRESS.md and log.txt) so a run can resume after interruption. ## Core Features & Use Cases - Subagent-driven execution: Each task is implemented by a fresh subagent with full context pasted in, while the controller session owns all git commits — subagents never run git add or git commit. - Risk-tiered review gates: Tasks touching types, data, or lib/ get two parallel review gates (spec compliance and code quality); pure UI/route wiring gets one combined gate, with every gate required to PASS before commit. - Resumable persistent state: PROGRESS.md checkboxes and an append-only log.txt live beside the plan, updated only after gates pass and the commit exists, making runs recoverable after session restarts. - Use Case: Given doc/features/005-record-expense/plan.md produced by the writing-plans skill, run /sdd to autonomously implement each numbered task, review it, commit it in conventional-commit format, and finish with a completion summary. ## Quick Start Run /sdd followed by the path to your plan file, for example /sdd doc/features/005-record-expense/plan.md, and let it execute each task through implementation, review, and commit.

Frequently Asked Questions about sdd

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I execute an implementation plan with subagents?

Run /sdd with the path to your plan.md file, for example /sdd doc/features/005-record-expense/plan.md. The skill dispatches a fresh implementer subagent per task, runs independent review gates, and commits only after every gate passes.

What is subagent-driven development for coding plans?

Subagent-driven development executes each plan task with a brand-new subagent that has no memory of prior tasks, while a controller session owns all git commits. Independent reviewer subagents verify spec compliance and code quality before anything is committed.

Can an interrupted plan execution be resumed?

Yes. The skill persists state in PROGRESS.md (one checkbox per task with commit hashes) and an append-only log.txt beside the plan file. On restart it reads PROGRESS.md as the source of truth and continues from the first unchecked task.

What happens when a review gate fails during execution?

A fresh fix subagent is dispatched scoped to only the failing findings, then the failed gates are re-run. If the same task fails a second consecutive time, execution stops and the user is asked whether to retry, skip, or take over manually.

When does a task get two review gates instead of one?

Tasks touching types, data, domain logic, lib/, or any module boundary get two independent gates (spec and code quality). Pure UI and route wiring with no lib/ change gets one combined gate, per the repo profile's gate risk tiers.