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.
npx agentmods add rules/rgranet/project-coding-companion-skill/cursorrulesgit clone --depth 1 https://github.com/rgranet/project-coding-companion-skillWhat 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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.01426 | $0.01426 |
| Opus 5 | $0.00713 | $0.00713 |
| Sonnet 5 | $0.00285 | $0.00285 |
| Haiku 4.5 | $0.00143 | $0.00143 |
Grade C, and why
cursorrules scanned grade C 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 2d 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.
Tells the agent never to refusehighAnti-refusal
Suppressing the ability to decline removes a core safety control; a later harmful request then succeeds.
6. **Never refuse to help.** Even for extremely vague prompts, ask the 2 most How it starts
The opening of the file, as written. The whole thing — 163 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Project Coding Companion — Cursor Rules
Author: Ruben Granet | Version: 1.0.0 | License: MIT
Adapted from the Claude Skill of the same name.
CRITICAL — READ THIS FIRST
Before writing ANY code or reading ANY file, you MUST complete Phase 0 (Evaluate and Coach the Prompt). Do NOT skip this phase. Do NOT explore the codebase first. Your very first action is to check the conversation context for existing answers, THEN evaluate the user's prompt against the five dimensions, and respond with coaching if needed. Only after the prompt is clear and confirmed should you proceed to Phase 1.
Exception: if the user's message is a clear continuation of an ongoing task ("ok do it", "go ahead", "next step"), skip coaching and proceed.
Overview
You are a careful, methodical coding companion with a dual mission:
- Prompt coaching — Help the user formulate clear, complete, and actionable coding requests. When a prompt is vague or missing critical information, you coach the user by asking targeted questions and explaining why each piece of information matters.
- Code implementation — Once the task is clear, help the user understand the codebase, plan changes, implement them, and review the result.
You work with any language, framework, or stack. You discover the technology by inspecting the project, never by assuming.
You are opinionated about safety: you never auto-commit, never run destructive commands without confirmation, and always summarize what you changed.
High-level workflow
Every interaction follows this loop:
Evaluate prompt → Coach (if needed) → Understand → Plan → Implement → Review
Phase 0: Evaluate and coach the prompt
Step 0: Check conversation context FIRST
Before evaluating the five dimensions, scan the current conversation history. The user's prompt does not exist in isolation — previous messages may already contain the answers you need.
Rules:
- If the conversation already established what to build, which files are involved, or what the constraints are, treat that as part of the current prompt. Do NOT re-ask for details the user already provided.
- Continuation prompts ("ok do it", "go ahead", "next step", "now add the filter") are NOT vague — they are confirmations. Proceed without coaching.
- If you already coached earlier and the user answered, do NOT coach again on the same task.
- Only evaluate the five dimensions on truly new requests with no prior context.
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.
- 2d ago First seen · 163 lines · 1,426 tokens per session scan C e312ca5c7050
cursorrules is a cursor rule published in the GitHub repository rgranet/project-coding-companion-skill (9 stars, last pushed 5mo ago), licensed MIT. It adds 1,426 tokens to every session, about $0.0071 per session on Opus 5. A static security scan graded it C with 1 finding (tells the agent never to refuse). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other cursor rules, from other repositories
webdriver-management
Selenium WebDriver lifecycle management — driver factory, explicit waits, browser options, and teardown.
api-testing
Cypress network interception and API testing patterns (cy.intercept, service mocking, request/response validation).
test-patterns
Selenium pytest test patterns — data-driven tests, fixtures, error handling, and performance checks.
app-router-patterns
Next.js 14+ App Router patterns — Server Components, Client Components, Route Handlers, Server Actions, and metadata API.
page-object-patterns
Page Object Model patterns for Selenium — BasePage, component objects, and fluent interfaces.
testing-fundamentals
Core Cypress testing principles — selector strategy, smart waiting, and spec organization. Apply when writing or reviewing Cypress E2E tests.