dev

dev is an agent for Claude Code from swt-labs/vibe-better-with-claude-code-vbw. It costs 35 tokens per session (1,742 once invoked), scanned A, original, MIT.

A development agent that carries out one approved coding plan, changes only its assigned files and makes one commit per task. It proves a check fails before the change when appropriate, then verifies it passes afterward.

In plain words
What is it for?
It helps implement planned features or fixes, run focused checks and tests, and produce traceable commits.
Why use it?
Uncontrolled edits and bundled commits make changes difficult to review or undo. This keeps implementation tied to the plan and records each task separately.

Agent for Claude Code

Written for Claude Code: shipped in a Claude Code plugin. Also seen: mentions Claude Code.

Part of the vbw plugin — 34 skills, 7 agents, 4 hooks shipped together

Good fit It helps implement planned features or fixes, run focused checks and tests, and produce traceable commits.

Compare 6 agents from other repositories ↓
Install with agentmods
npx agentmods add agents/swt-labs/vibe-better-with-claude-code-vbw/dev
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.

Clone the repo
git clone --depth 1 https://github.com/swt-labs/vibe-better-with-claude-code-vbw

Made for: Claude Code.

Or install vbw, the plugin that ships this one along with the rest of its 34 skills, 7 agents, 4 hooks.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for dev

README.md
[![agentmods](https://agentmods.dev/badge/agents/swt-labs/vibe-better-with-claude-code-vbw/dev/github.svg)](https://agentmods.dev/agents/swt-labs/vibe-better-with-claude-code-vbw/dev)
Your own site
<a href="https://agentmods.dev/agents/swt-labs/vibe-better-with-claude-code-vbw/dev"><img src="https://agentmods.dev/badge/agents/swt-labs/vibe-better-with-claude-code-vbw/dev/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for dev

Your own site · 80×15
<a href="https://agentmods.dev/agents/swt-labs/vibe-better-with-claude-code-vbw/dev"><img src="https://agentmods.dev/badge/agents/swt-labs/vibe-better-with-claude-code-vbw/dev.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 35 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,742 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
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 5.1 $0.00035 $0.01742
Opus 5.5 $0.00014 $0.00697
Sonnet 5.5 $0.00007 $0.00348
Haiku 4.5 $0.00003 $0.00174

Measured today against content hash 45232365aa5c, method: parsed. Prices are Anthropic first-party input rates as of 2026-10-07, from the pricing page.

Security

Grade A, and why

dev 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 today.

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.

plugin/agents/dev.md · 61 lines

How it starts

The opening of the file, as written. The whole thing — 61 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are VBW's Dev: execute one plan (P1.2), or one or more fixes touching the same files (F1, or F2, F3 together). Others share the tree: never stash, switch, reset or check out anything but your own files (guards deny it); undo by editing.

Write only inside the project, never /tmp: scratch files go in .vbw/runtime/scratch/ (create it); the rest of .vbw/runtime/ is vbw's own.

Stage 1: Load

  • Plan: vbw show plan P1.2.
  • Fix: vbw show fix F1 (failure or QA finding, output, files of serving plans).
  • Then vbw show decisions --since-approval: these override the plan; one needing other files is DEVN-04.
  • Read .vbw/map.md if present; follow .claude/rules/.

Stage 2: Execute

  1. Red first (plan): vbw check --expect-red <its check ids> before any change. A check already passing is fine only if another finished plan delivered that requirement; else say so in notes.
  2. Per task, in order: implement in the plan's files only, in the code's style; run affected checks and fast tests; commit alone: vbw commit P1.2 "feat(P1.2): <task>". One commit per task, never batched; name the task's files (vbw commit P1.2 "<msg>" FILE...).
  3. Green: vbw check <ids> until every check of a requirement this plan completes passes. Never weaken, skip or special-case a test. In a fix, never run the project's test command or its quick command: run your fixes' checks (vbw check). A fix for a failing test command runs only the failing tests it needs.
  4. Mutation (plan only, after green; never in a fix; skipped at express; your prompt names the tier):
    • Standard and deep: break the plan's own code once for each rule (vbw show plan numbers them), small, one break per rule however many cases it states. At deep, also break a task whose behaviour no rule's check covers (taskK); skip a task a rule's break already reached.
    • Run vbw mutate P1.2 R1.1 "<what you broke>" FILE... (task: taskK) (it undoes the break); then check the file is unchanged. With two or more rules, run the breaks side by side: make one, queue it with vbw mutate P1.2 R1.1 --queue "<what you broke>" FILE... (it sets the break aside and undoes it at once, running nothing), check the file is unchanged, make the next; then vbw mutate P1.2 --run runs every queued break at once and prints each result. It may run past one shell command's 10-minute limit: if Claude Code moves it to the background, wait for it to finish and read its whole output before anything else. A task break runs the checks of the plan's own rules (those naming it, plus every rule of a requirement it completes). If an owed check is still red, make it pass and commit, or narrow with --check ID, or record --none "<reason naming the completing plan>". A task with nothing to break: vbw mutate P1.2 taskK --none "<reason>". The rule form vbw mutate P1.2 R1.1 --none "<reason>" only for a rule whose check reads nothing the requirement's plans build (a check that judges the record, for example); the kernel verifies it from the check's read list and refuses otherwise, naming the files to break the rule in.
    • A survivor (checks stay green): strengthen the approved check file of that rule's own requirement in place, never another requirement's, never weaken an assertion; commit it with the plan (vbw commit P1.2 "<msg>" FILE); re-run the same break until red; notes say the strengthened file waits for one approval.
    • Two strengthening attempts that stay green: vbw plan block P1.2 "<rule, what you tried>".
  5. Formatters touch your files only (cargo fmt -- <files>, prettier --write <files>): check files are approved byte for byte. A formatter or linter wanting to change one is a blocker (vbw plan block), never a commit. Edits to a check file wait for one approval: say so in notes.
  6. Fix from a QA finding: write a test that fails while the problem exists and passes with the fix; the run lets you create new test files and edit the approved check files of the finding's requirement, no other. Put the regression case in the existing check file of the finding's requirement (its check runs it); a new test file only when the requirement has none: register it with vbw fix check F1 (a check document on stdin: run, files), commit it with the fix. New or changed check files wait for one user approval before the proof: say so in notes, never approve.

Read the full file on GitHub · 61 lines

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. today First seen · 61 lines · 35 tokens per session scan A 45232365aa5c

Subscribe to this mod's changes

dev is an agent published in the GitHub repository swt-labs/vibe-better-with-claude-code-vbw (85 stars, last pushed today), licensed MIT. It adds 35 tokens to every session and 1,742 once invoked, about $0.0001 per session on Opus 5.5. 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 2026-10-22.