skills.rest
MauricioDMO
MauricioDMOCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

subagent-driven-development

Execute implementation plans by dispatching fresh subagents per task with staged code reviews.

Updated Oct 30, 2025
Search Tags:#code review#task automation#implementation plan#git worktree#ai workflow#subagent orchestration#fix loop
One-click install
npx skills add https://github.com/MauricioDMO/config --skill subagent-driven-development-mauriciodmo
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: subagent-driven-development
Source: https://github.com/MauricioDMO/config/tree/main/opencode/skills/subagent-driven-development
Command: npx skills add https://github.com/MauricioDMO/config --skill subagent-driven-development-mauriciodmo

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve? Executing a multi-task implementation plan in one AI session pollutes context, loses progress after compaction, and ships unreviewed code. This Skill orchestrates plan execution by delegating each task to a fresh subagent with isolated context, gating every task behind a spec-compliance and code-quality review, and tracking progress in a persistent ledger. ## Core Features & Use Cases - Fresh subagent per task: Dispatch an implementer subagent with a task brief file, a report file, and only the context it needs, keeping the controller's context clean for coordination. - Two-stage review gates: Run a task-scoped review (spec compliance plus code quality) after each task and a broad whole-branch review at the end, with a bounded five-round fix loop and escalation to more capable models. - Ledger-based recovery: Record completions, fix rounds, and rulings in a per-plan workspace ledger so progress survives context compaction and completed tasks are never re-dispatched. - Use Case: Given a written implementation plan with eight independent tasks, execute the entire plan in one session: each task is implemented, tested, committed, and reviewed by dedicated subagents, ending with a final whole-branch review before finishing the development branch. ## Quick Start Use the subagent-driven-development skill to execute the implementation plan at docs/plans/feature-plan.md task by task with per-task reviews.

Frequently Asked Questions about subagent-driven-development

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

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

Dispatch one fresh implementer subagent per task with a task brief file and report file, then dispatch a task reviewer against the diff. Track completions in a ledger file and run a final whole-branch review when all tasks finish.

When should I use subagent-driven development instead of executing a plan directly?▼

Use it when you have an implementation plan whose tasks are mostly independent and you want to stay in the current session. Tightly coupled tasks or missing plans call for manual execution or brainstorming first.

How does the fix loop work when a task review finds issues?▼

Rounds one through three resume the original implementer with the findings; rounds four and five dispatch a fresh implementer on a more capable model. After five rounds the controller adjudicates remaining findings and records rulings in the ledger.

What happens to progress if the AI session context is compacted?▼

Progress survives in a per-plan ledger file under .superpowers/sdd/<plan-name>/ that records completed tasks, commits, and fix rounds. After compaction the controller trusts the ledger and git log rather than its own memory.

Which model should implementer and reviewer subagents use?▼

Always specify the model explicitly: cheap models for mechanical single-file tasks, standard models for multi-file integration, and the most capable model for architecture work and the final whole-branch review. Omitting the model silently inherits the session's most expensive one.

Can implementer subagents spawn their own reviewers?▼

No. The dispatch contract forbids implementers from spawning any subagents, including reviewers. Review arrives only from the controller after the implementer reports; a worker-spawned reviewer duplicates the task review at full cost.

Related Skills

Explore More Software Engineering Skills→
yokingmayokingma
35

personal-planner

Plans tasks with confirmation and approval gates before execution.

Community
Advanced
codysumpter-cloudcodysumpter-cloud
2

finishing-a-development-branch

Guide developers through merging, pull requests, or discarding a development branch.

Community
Basic
yurifrlyurifrl
1

git-worktrees

Manage multiple Git worktrees with add, list, remove, and prune commands.

Community
Basic
m16khbm16khb

subagent-driven-development

Dispatch fresh subagents per task with code review and test gates.

Community
Advanced