subagent-driven-development

A development workflow that assigns each task in an implementation plan to a fresh coding subagent, then sends the result through several reviews.

In plain words
What is it for?
Use it for plans with at least two mostly independent tasks that can be handled in parallel while staying in the same session.
Why use it?
It separates tasks and review responsibilities, helping catch specification, code-quality, architecture, testing, and sometimes security problems before the work is considered complete.

Skill for Claude CodeCodex

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add skills/aram-devdocs/plumb/subagent-driven-development
Any agent
npx skills add aram-devdocs/plumb --skill subagent-driven-development
Clone the repo
git clone --depth 1 https://github.com/aram-devdocs/plumb

Made for: Claude Code, Codex.

Per session 59 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 2,627 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00059 $0.02627
Opus 5 $0.00030 $0.01314
Sonnet 5 $0.00012 $0.00525
Haiku 4.5 $0.00006 $0.00263

Measured yesterday against content hash 01145a7ff726, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

subagent-driven-development scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.agents/skills/subagent-driven-development/SKILL.md · 255 lines

How it starts

The opening of the file, as written. The whole thing — 255 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Subagent-driven development

Execute a plan by dispatching a fresh subagent per task. After each task, run the Plumb review-gate chain. For independent tasks, dispatch in parallel batches for maximum throughput.

Core principle: fresh subagent per task + parallel batches for independent work + ordered review gates = high quality, fast iteration.

When to use

  • You have a plan with ≥2 tasks.
  • Tasks are mostly independent (different crates, no shared edits, no producer-consumer edges).
  • You want to stay in the current session (no context switch).

If tasks are tightly coupled or the plan needs brainstorming, execute manually or use /gh-issue with a single primary.

Process overview

Per task:
  Dispatch 01-implementer → self-review → dispatch 02-spec-reviewer → (fixes loop)
    → dispatch 03-code-quality-reviewer → (fixes loop)
    → dispatch 05-architecture-validator → (fixes loop)
    → dispatch 04-test-runner → (fixes loop)
    → (if CDP/MCP/URL/dep touched) dispatch 06-security-auditor in parallel with any of the above
  Mark task complete.

After all tasks:
  Dispatch a final 04-test-runner pass on the full changeset.
  Run `just validate` + `cargo xtask pre-release`.

Parallel batch workflow

Step 1 — task decomposition

For each task, identify:

  • Target files — which files will be created / modified.
  • Target crates — which of plumb-core, plumb-format, plumb-cdp, plumb-config, plumb-mcp, plumb-cli, xtask are touched.
  • Dependencies — what public items does the task import?
  • Outputs — what does the task produce that another might consume (new public API, new MCP tool schema, new config field)?

Step 2 — dependency analysis

Classify each task pair as INDEPENDENT or DEPENDENT.

INDEPENDENT (safe to parallelize) — every one of these must hold:

  • Different target files (no shared edits).
  • Different crates OR no upward dependency change.
  • No change to a shared public API consumed by another task.
  • No producer-consumer relationship (one task's public output isn't another's input).
  • Can be merged without conflicts.

Read the full file on GitHub · 255 lines

Files

What ships with it

3 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. yesterday First seen · 255 lines · 59 tokens per session scan A 01145a7ff726

Subscribe to this mod's changes

subagent-driven-development is a skill published in the GitHub repository aram-devdocs/plumb (2 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 59 tokens to every session and 2,627 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

Other skills, from other repositories

extract-design

Extract the full design language from any website URL. Produces 8 output files including AI-optimized markdown, visual HTML preview, Tailwind config, React theme, shadcn/ui theme, Figma variables, W3C design tokens, and CSS variables. Also runs WCAG accessibility scoring. Use when user says 'extract design', 'get…

Manavarya09/design-extract · 98 tokens

designlang-tokens

Use when styling UI for cal.com — references the extracted design system tokens instead of inventing colors, spacing, or typography.

Manavarya09/design-extract · 30 tokens

uisandbox

Open a real, built web app in UISandbox — the interactive sandbox in the user's browser, 1:1, with the app's own knobs (brand, colours, fonts, radius, spacing…) — or test one change headlessly, verify it still renders 1:1, get a screenshot and the patch. Use when someone wants to look at their app in a sandbox and…

Ideelab/uisandbox · 155 tokens

design-md-extractor

Extract and document the visual style of public websites as reusable DESIGN.md files. Use when the user asks to analyze a website's design system, extract styles, create design tokens, produce DESIGN.md, generate an agent-readable style guide, or build from a website's visual language.

jpoindexter/design-md-extractor · 60 tokens

ss-score

Score a visual artifact's implementation quality 0-100 against its composed StyleSeed rule set — category breakdown, evidence, and prioritized fixes.

bitjaru/styleseed · 31 tokens

ss-setup

Configure StyleSeed by selecting the output grammar, domain, page type, brand recipe, semantic palette recipe, optional aesthetic profile, and bounded brand tokens before scaffolding a first screen.

bitjaru/styleseed · 40 tokens