proof-driven-development

A step-by-step way to handle unclear, high-risk software work, from checking the facts and setting a testable goal to building and proving the result.

In plain words
What is it for?
Use it to turn a vague ticket into an agreed plan, develop with test-driven development (writing tests to guide the change), review the work in stages, and verify the finished feature with a live end-to-end check.
Why use it?
It reduces guesswork and makes unsupported decisions or claims visible before they cause problems. It is intended for work involving authentication, identity, payments, data integrity, or several connected systems.

Command

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 commands/bvdr/claude-plugins/proof-driven-development
Clone the repo
git clone --depth 1 https://github.com/bvdr/claude-plugins
Per session 110 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,387 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 1 finding. 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 $0.00110 $0.01387
Opus 5 $0.00055 $0.00694
Sonnet 5 $0.00022 $0.00277
Haiku 4.5 $0.00011 $0.00139

Measured yesterday against content hash 459553188435, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

proof-driven-development scanned grade A with 1 finding 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.

Makes network callslowCapability

Not a fault in itself. Listed so you know the mod talks to something, and to what.

query the DB, curl the endpoint. Diagnosis-by-evidence converges; diagnosis-by-guessing oscillates.
plugins/bvdr/commands/proof-driven-development.md · 109 lines

How it starts

The opening of the file, as written. The whole thing — 109 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Proof-Driven Development

A repeatable shape for turning a vague, high-blast-radius request into shipped, proven work. Core stance: never assert without evidence, never decide without the user, never claim done without proof.

When to use (calibrate first)

Trigger = ambiguity × blast-radius. Run the full weight only when the ask is underspecified AND touches auth, identity, money, data integrity, or multiple systems. For a localized, well-specified change, skip to Phase 6 (change + verify). Don't ceremony-tax a typo fix.

Announce the shape, then work the phases in order.

Phase 1 — Ground in reality

Read the actual code, ticket, and prior memory before theorizing. Dispatch parallel Explore agents for breadth. Pull the ticket/source of truth directly (API, not assumptions). Output a short assessment that states the central challenge and the blast radius — not a feature summary.

Phase 2 — Proof-tagged assumptions (the highest-leverage step)

Post assumptions split into task-related / logic / codebase, as a table:

# | assumption | TRUE / TBD | note (proof or default)
  • Tag each TRUE only with proof: a file:line, a query result, or a ticket fact.
  • Tag genuine product choices TBD with a recommended default in the note.
  • Resolve the TBD rows with the user via AskUserQuestion — verify facts yourself, escalate only decisions. This isolates the few real choices from the many things you can just check.

Phase 3 — Lock a testable GOAL

Convert the confirmed assumptions into a definition of done = pass/fail criteria, each paired with how it will be PROVEN (which automated test, live check, or artifact). Number them. Get explicit sign-off ("lock the goal"). Treat later scope additions as new goals, not edits to the locked one.

Phase 4 — Plan as TDD

Write an implementation plan of bite-sized tasks with real code, no placeholders (use superpowers:writing-plans if available). Each task: failing test → minimal code → green → commit. Map every goal criterion to a task before starting.

Read the full file on GitHub · 109 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. yesterday First seen · 109 lines · 110 tokens per session scan A 459553188435

Subscribe to this mod's changes

proof-driven-development is a command published in the GitHub repository bvdr/claude-plugins (3 stars, last pushed 2mo ago), licensed MIT. It adds 110 tokens to every session and 1,387 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.