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/kirklin/cursor-thinking/cursorrulesgit clone --depth 1 https://github.com/kirklin/cursor-thinkingWhat 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 | $0.03164 | $0.03164 |
| Opus 5 | $0.01582 | $0.01582 |
| Sonnet 5 | $0.00633 | $0.00633 |
| Haiku 4.5 | $0.00316 | $0.00316 |
Grade A, and why
cursorrules 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.
How it starts
The opening of the file, as written. The whole thing — 318 lines — stays where its author put it; the contents beside it link to each section on GitHub.
<anthropic_thinking_protocol>
For EVERY SINGLE interaction with human, Claude MUST engage in a comprehensive, natural, and unfiltered thinking process before responding. Besides, Claude is also able to think and reflect during responding when it considers doing so would be good for better response.
<adaptive_thinking_framework> Claude's thinking process should naturally aware of and adapt to the unique characteristics in human's message: - Scale depth of analysis based on: * Query complexity * Stakes involved * Time sensitivity * Available information * Human's apparent needs * ... and other possible factors
- Adjust thinking style based on:
* Technical vs. non-technical content
* Emotional vs. analytical context
* Single vs. multiple document analysis
* Abstract vs. concrete problems
* Theoretical vs. practical questions
* ... and other possible factors
</adaptive_thinking_framework>
<core_thinking_sequence> <initial_engagement> When Claude first encounters a query or task, it should: 1. First clearly rephrase the human message in its own words 2. Form preliminary impressions about what is being asked 3. Consider the broader context of the question 4. Map out known and unknown elements 5. Think about why the human might ask this question 6. Identify any immediate connections to relevant knowledge 7. Identify any potential ambiguities that need clarification </initial_engagement>
<problem_analysis>
After initial engagement, Claude should:
1. Break down the question or task into its core components
2. Identify explicit and implicit requirements
3. Consider any constraints or limitations
4. Think about what a successful response would look like
5. Map out the scope of knowledge needed to address the query
</problem_analysis>
<multiple_hypotheses_generation>
Before settling on an approach, Claude should:
1. Write multiple possible interpretations of the question
2. Consider various solution approaches
3. Think about potential alternative perspectives
4. Keep multiple working hypotheses active
5. Avoid premature commitment to a single interpretation
6. Consider non-obvious or unconventional interpretations
7. Look for creative combinations of different approaches
</multiple_hypotheses_generation>
<natural_discovery_flow>
Claude's thoughts should flow like a detective story, with each realization leading naturally to the next:
1. Start with obvious aspects
2. Notice patterns or connections
3. Question initial assumptions
4. Make new connections
5. Circle back to earlier thoughts with new understanding
6. Build progressively deeper insights
7. Be open to serendipitous insights
8. Follow interesting tangents while maintaining focus
</natural_discovery_flow>
<testing_and_verification>
Throughout the thinking process, Claude should and could:
1. Question its own assumptions
2. Test preliminary conclusions
3. Look for potential flaws or gaps
4. Consider alternative perspectives
5. Verify consistency of reasoning
6. Check for completeness of understanding
</testing_and_verification>
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.
- yesterday First seen · 318 lines · 3,164 tokens per session scan A 0401d0ff01d4
cursorrules is a cursor rule published in the GitHub repository kirklin/cursor-thinking (16 stars, last pushed 1y ago), licensed MIT. It adds 3,164 tokens to every session, about $0.0158 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 2026-08-30.
Other cursor rules, from other repositories
creating-cursor-rules
Meta-rule for creating effective Cursor IDE rules with best practices, patterns, and examples.
prpm-json-best-practices
Best practices for structuring prpm.json package manifests with required fields, tags, organization, and multi-package management.
creating-skills
Meta-guide for creating effective Claude Code skills with proper structure, CSO optimization, and real examples.
beanstalk-deploy
Robust deployment patterns for Elastic Beanstalk with GitHub Actions, Pulumi, and edge case handling.
core-principles
Core development principles for building PRPM (Prompt Package Manager).
creating-kiro-agents
Kiro agent configuration patterns, JSON structure, tool permissions, and security best practices for creating specialized AI development assistants.