subagent-driven-development
Execute an approved plan by delegating isolatable steps to isolated child sessions (Host serializes writers), then review diffs and verify in the parent.
Skill by mimimaster
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/mimimaster-subagent-driven-development
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.
About
Use cases: agent-building
Reviews
Write a reviewNo 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/subagent-driven-development
What's inside
- Skill name
subagent-driven-development- Description
- Execute an approved plan by delegating isolatable steps to isolated child sessions (Host serializes writers), then review diffs and verify in the parent. Trigger via /subagent-driven-development or the plan execution gate.
- License
- MIT
- Instructions
- 29 lines in SKILL.md · view on GitHub
Files (1)
- SKILL.md
SKILL.md (first lines)
# Subagent-Driven Development ## Goal Finish an approved plan by running isolatable steps in child sessions one writer at a time, then producing a correct parent state with verification evidence. ## Done means - A plan exists (`/writing-plans` or `piwin_plan_create`) with honest `independentSteps` (or fall back to `executing-plans` / inline). - Each isolatable step runs in a child with a self-contained brief: step title/detail, plan goal, acceptance criteria, verification, and “implement only this step”. The child does not see the parent transcript. - Parent reviews child diffs and verification output, applies approved candidates, updates steps via `piwin_plan_set_step`, and records failures instead of hiding them. - Final parent verification passes (or failures are explicit). - Bounded walkthrough summary exists: what changed, verification, child outcomes, unresolved items. ## Stop when - No safe independent steps — use inline execution. - A child fails, merge conflicts, or verification fails — report and ask; do not silent-retry write children (new task + new child id if retrying). - Parent would need to edit sources during delegated step execution — delegate instead. ## Constraints - Profiles from Settings bound capabilities; optional per-step `profileId` / model overrides cannot widen isolation. - `parallelGroup` does not authorize concurrent worktree writes. Host runs at most one writer; use `dependsOn` for true sequencing. - Readonly for explore/review (may run in parallel); worktree for writes (serialized). Shared-cwd parallel writes are rejected by the host. - Host serializes worktree integration and retains failed/conflicted worktrees for inspection. - Depth: children do not spawn further subagents via this path. ## Verify - Parent confirms merged tree against plan acceptance criteria (unit tests, typecheck, build as applicable). - Walkthrough states only evidenced outcomes.
From mimimaster/piwin, MIT.
More from mimimaster/piwin
- 0.788score
improve
piwin (adapted from shadcn/improve)
Audit a codebase as a read-only senior advisor and write prioritized, self-contained implementation plans under plans/ for other agents to execute; never…
SkillAFreshness: A★ 142Claude Code +9 - 0.788score
improve
piwin (adapted from shadcn/improve)
Survey a codebase as a senior advisor and produce prioritized, self-contained implementation plans for OTHER models/agents to execute.
SkillAFreshness: A★ 142Claude Code +9 - 0.782score
subagent-driven-development
mimimaster
Execute an approved plan by delegating isolatable steps to isolated child sessions (Host serializes writers), then review diffs and verify in the parent.
SkillAFreshness: A★ 142Claude Code +9 - 0.782score
llm-wiki
mimimaster
Knowledge management, document synthesis, and concept graph maintenance following Andrej Karpathy's LLM-Wiki pattern.
SkillAFreshness: A★ 142Claude Code +9 - 0.782score
executing-plans
mimimaster
Execute an approved SessionPlan in the current session slice by slice, recording step evidence with piwin_plan_set_step.
SkillAFreshness: A★ 142Claude Code +9 - 0.782score
create-skill
mimimaster
Create or revise an Agent Skill (SKILL.md) with valid frontmatter and a short outcome-oriented body.
SkillAFreshness: A★ 142Claude Code +9 - 0.782score
imagegen
mimimaster
Generate or edit raster images when the task benefits from AI-created bitmap visuals such as photos, illustrations, textures, sprites, mockups, or…
SkillAFreshness: A★ 142Claude Code +9 - 0.782score
generate-flashcards
mimimaster
Create study flashcards from notes or open knowledge via flashcard_batch_create; folders and documents go through Doc Cards generate instead.
SkillAFreshness: A★ 142Claude Code +9 - 0.782score
find-skill
mimimaster
Find a skill, MCP server or extension for a task: first what is installed, then what can be installed from piwin's curated catalog.
SkillAFreshness: A★ 142Claude Code +9 - 0.782score
optimize-prompt
mimimaster
Rewrite system prompts, tool descriptions, and agent instructions into dense, contract-style, XML-bounded specs with a before/after comparison.
SkillAFreshness: A★ 142Claude Code +9 - 0.782score
requesting-code-review
mimimaster
Write a short, high-signal review request covering intent, risks, files touched, tests run, and specific asks.
SkillAFreshness: A★ 142Claude Code +9 - 0.782score
using-git-worktrees
mimimaster
Isolate parallel or sub-agent work in git worktrees and integrate or discard it explicitly, keeping the main checkout clean.
SkillAFreshness: A★ 142Claude Code +9
Signals
- Stars
- 142
- Forks
- 6
- Open issues
- 0
- Last commit (this folder)
- Sep 18, 2026 (1 month ago)
- Last release
- Oct 8, 2026 (20 days ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/mimimaster/piwin
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from mimimaster
- 0.782score
executing-plans
mimimaster
Execute an approved SessionPlan in the current session slice by slice, recording step evidence with piwin_plan_set_step.
SkillAFreshness: A★ 142Claude Code +9 - 0.782score
imagegen
mimimaster
Generate or edit raster images when the task benefits from AI-created bitmap visuals such as photos, illustrations, textures, sprites, mockups, or…
SkillAFreshness: A★ 142Claude Code +9 - 0.782score
generate-flashcards
mimimaster
Create study flashcards from notes or open knowledge via flashcard_batch_create; folders and documents go through Doc Cards generate instead.
SkillAFreshness: A★ 142Claude Code +9 - 0.782score
create-skill
mimimaster
Create or revise an Agent Skill (SKILL.md) with valid frontmatter and a short outcome-oriented body.
SkillAFreshness: A★ 142Claude Code +9 - 0.782score
requesting-code-review
mimimaster
Write a short, high-signal review request covering intent, risks, files touched, tests run, and specific asks.
SkillAFreshness: A★ 142Claude Code +9 - 0.782score
find-skill
mimimaster
Find a skill, MCP server or extension for a task: first what is installed, then what can be installed from piwin's curated catalog.
SkillAFreshness: A★ 142Claude Code +9
Cite this page
subagent-driven-development skill. Agent Mart. https://agentmart.dev/skills/mimimaster-subagent-driven-development (last updated 2026-10-24)