proofloop CLAUDE.md

A set of project instructions for Proofloop, a Claude Code and Claude Cowork plugin that checks how well agent tasks are carried out.

In plain words
What is it for?
Use it when developing, testing, or reviewing Proofloop's quality scoring, rubrics, scorecards, hooks, or manual judging command.
Why use it?
It defines the supported scoring areas, review rules, and operating boundaries so evaluations are consistent and do not claim unsupported coverage.

Instructions file

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 instructions/sattyamjjain/proofloop/claude-md
Clone the repo
git clone --depth 1 https://github.com/sattyamjjain/proofloop
Per session 4,440 This file is loaded in full into every session.
When invoked 4,440 The same file — it is already loaded in full.
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 $0.04440 $0.04440
Opus 5 $0.02220 $0.02220
Sonnet 5 $0.00888 $0.00888
Haiku 4.5 $0.00444 $0.00444

Measured 3d ago against content hash e2d61ecea01c, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

proofloop CLAUDE.md 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 3d ago.

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.

This project has no build step or package manager. Python scripts use stdlib only (Python 3.9+); the optional LLM second-opinion analyzer uses `urllib.request` rather than pulling in the `anthropic` SDK.
CLAUDE.md · 240 lines

How it starts

The opening of the file, as written. The whole thing — 240 lines — stays where its author put it; the contents beside it link to each section on GitHub.

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Overview

Proofloop is a Claude Code plugin that auto-evaluates skill and agent execution quality. It provides 7-dimension scoring (correctness, completeness, adherence, actionability, efficiency, safety, consistency), configurable rubrics, persistent scorecards, and dual-mode operation (auto via hooks + manual via /judge command). Works on both Claude Code and Claude Cowork.

  • Author: Sattyam Jain
  • License: MIT
  • Platforms: claude-code, claude-cowork

v4.3 Scope Contract (2026-05-03)

Proofloop is a Claude Code / Cowork plugin only. Per the 2026-05-03 v4.3 scope reset (runbook §scope-reset), the rubric and adapter inventory is pinned to plugin-domain quality scoring. Frontier-lab eval-bench scope (SWE-bench, Terminal-Bench, GAIA, OSWorld, MCP attack benches, etc.) is explicitly out of scope and will not be re-added without a runbook spec change.

In-scope rubrics (11): code-review, security, devops, data-analysis, frontend-design, testing, documentation, content-writing, research, default, custom-template.

Out-of-scope, queued for v2.0.0 trim (16): agentic-sast-confidence, browser-agent, clinical-agentic-workflow, code-review-aider-polyglot, eu-ai-act-audit-trail, function-hijacking-robustness, gpt-5-5-differential, model-spec-compliance, owasp-mcp-top-10-beta, project-deal-commerce, routine-execution, ship-readiness, skill-compliance, swe-bench-pro, terminal-bench, tool-output-rewrite.

In-scope adapters (4): claude_code, cowork, openai_compatible, codex. Cross-ecosystem adapters (Gemini, MLflow, Inspect-AI, Terminal-Bench, browser-harness) are out of scope.

Core surfaces (do not regress): 7-dimension scoring (correctness, completeness, adherence, actionability, efficiency, safety, consistency); auto-hooks (Stop, SubagentStop, StopFailure); /judge, /scorecard, /benchmark, /judge-config, /against, /compare slash commands; per-rubric <name>.weights.json sidecar overrides; threshold blocking via judge-config.json.

Read the full file on GitHub · 240 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. 3d ago First seen · 240 lines · 4,440 tokens per session scan A e2d61ecea01c

Subscribe to this mod's changes

proofloop CLAUDE.md is an instructions file published in the GitHub repository sattyamjjain/proofloop (5 stars, last pushed 2mo ago), licensed MIT. It adds 4,440 tokens to every session, about $0.0222 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.