mirror of https://github.com/N1-AI/claude-hidden-toolkit.git synced 2026-06-23 11:39:56 +02:00

Files

T

frmorettoandClaude Opus 4.6 a7197e5534 Edition 1.4: 37 tools, 4 platforms, 13 corrections, 2 contributors

9 new tool cards (29–37): iOS Reminders CRUD suite (5 tools),
visualize:show_widget, gmail_create_draft, anthropic_api_in_artifacts,
persistent_storage. 13 v1.3 corrections including user_time/location
reclassified as always-loaded, memory_user_edits 500-char limit,
chart_display_v0 intermittent (not deterministic), window.storage
session-scoped. New sections: MCP connector state, artifact execution
layer, Project context behavior. iOS split from Android as distinct
platform. Contributors: anonymous security researcher, DMontgomery40.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

2026-03-17 19:29:02 +01:00

2.1 KiB

Raw Permalink Blame History

Security Policy

Scope

This project documents Claude.ai's internal tools through reverse-engineering. While the book itself is documentation (not executable code), the subject matter touches on AI system internals, and discoveries may occasionally surface behaviors with security implications.

Reporting a Vulnerability

If you discover a tool behavior that could be exploited to:

Bypass Claude's safety guardrails
Access other users' data through shared storage or memory tools
Exfiltrate sensitive information via tool parameters or responses
Manipulate tool behavior through prompt injection in tool inputs
Escalate permissions beyond intended scope
Abuse the artifact execution layer (anthropic_api_in_artifacts, persistent_storage, MCP endpoints) for unintended access

Please report it responsibly:

Do NOT open a public Issue. Security-relevant findings should not be disclosed publicly before they can be addressed.
Email: Contact the maintainer directly at the email listed on the author's GitHub profile.
Include:
- Description of the behavior
- Steps to reproduce
- Platform(s) affected
- Potential impact assessment
Also consider reporting to Anthropic directly via their responsible disclosure process — they are best positioned to fix tool-level vulnerabilities.

What This Project Will Do

Acknowledge receipt within 48 hours
Assess whether the finding should be withheld from publication until Anthropic addresses it
Coordinate with Anthropic if appropriate
Credit the reporter (unless anonymity is requested) when the finding is eventually published

What This Project Will NOT Do

Publish exploit details before Anthropic has had reasonable time to respond
Document tools or behaviors specifically designed to facilitate harm
Provide instructions for bypassing safety systems

Supported Versions

Version	Supported
1.4	Yes
1.3	No — please reference the latest edition
< 1.3	No

2.1 KiB Raw Permalink Blame History