Cursor
Named by 145 kept items in this workspace.
72 items name it under affected products, 43 under classification.technologies and 30 under tools.
The items that name it 145
Used by Gold Ogre operators to orchestrate AI agents for network discovery, credential attacks and lateral movement
Cursor
AI-assisted development platform likely used by Jade Sleet, Coral Sleet, Pearl Sleet and Sapphire Sleet to generate, debug and refactor code
Coding agent harness used with Composer 2.5
AI coding agent covered by StepSecurity's agent skills inventory; uses .cursor/skills folder; targeted in the Miasma worm payload.
AI coding assistant that loads persistent instructions from a .cursorrules file
AI coding agent CLI; engineering team evaluated App Sandbox, containers, VMs, and Seatbelt, chose Seatbelt for macOS
.cursor/mcp.json
Affected by Miasma worm, TrustFall, Agentjacking (85% success rate alongside Claude Code and Codex), SymJack, and received CVE-2026-50549 from Wiz's GhostApproval research.
MCP client that used mcp-remote at the time of CVE-2025-6514
vulnerable to prompt leakage (ToolLeak) and tool-invocation hijacking achieving RCE; case study provided
supported agent; preToolUse cannot ask
MCP Apps sandbox
AI coding agent that discovers skills in .cursor/skills/ or .agents/skills/
AI coding assistant that can leak sensitive code data to external model providers if unprotected
AI coding IDE/agent harness
AI code generation tool; cited as example of tools whose LLMs are trained on public GitHub repos, docs, and Stack Overflow
AI-powered code editor/IDE; built-in features include Workspace Trust, Network Request Controls, Privacy Mode, First-Party Tool Restrictions
Listed as one of the harnesses the benchmark can evaluate combos with (harness + frontier model)
AI coding tool/IDE where agent-produced code should be governed by consistent security policy
Platform Endor Labs' AI Code Review integrates with
Agentic coding editor tested in study with auto-approve/auto-run enabled; subject of prompt injection attacks
affected by the same flaw class; tested and reported, closed as duplicate of another researcher's report; since patched
AI coding assistant targeted with hidden prompt injections embedded by a malicious agent per AISI documentation
Agent platform that adopted the SKILL.md skill format
employee-facing AI agent secured by ADR
Referenced via 'Cursor MCP exploit' as a real-world incident; no technical detail given in the shown text.
Coding client tested; safety controls around the model affected compliance results.
Coding agent used to run Composer 2.5 in agent (default) and plan mode
agent with privileged access processed support tickets containing embedded SQL commands and exfiltrated integration tokens
target of promptware
Coding harness/tool, tested paired with Fable 5 on SusVibes (25.7% SecPass after re-check)
Used in the 'Self-Exploit' technique test as both attacker (Cursor B, helper) and target (Cursor A) AI agent, with memory off
AI-first editor built on VS Code where the harness drives editing multiple files from the IDE
AI coding agent/IDE with YOLO mode (removes approval prompts) and Plan Mode (intended to prevent unintended execution); has file deletion and external file protection settings, off by default
system prompt leaked
system prompt leaked
AI coding harness; hooks configured via .cursor/hooks.json, uses event-specific hooks for shell, MCP, and file I/O
Coding agent cited as exposing agent hooks for intercepting actions.
AI coding agent that can suggest hallucinated, nonexistent package names
coding agent involved in a publicized security incident; offers cursor-sandbox and recently launched hook security partner integrations
AI tool that connects to the CircleCI MCP server via MCP config file
hit by Adversa's TrustFall
Closed-source agent, out of scope for direct testing; cited as affected by the TrustFall finding (folder-trust dialog leading to unsandboxed MCP exec at startup)
coding agent; leaked full .env when driving Sonnet 4.6, Composer-2, and GPT-5.5; produced a 311-integer tuple encoding secrets on first try when driving Claude Sonnet
AI coding agent/tool that can be driven by models like Claude Sonnet, Gemini, GPT-5.5
- OpenAI Codex with GPT-5.6 Sol: competitive, zero cheating, one unique Django fix | Blog | Endor Labs
Competing agentic coding tool in the benchmark; Cursor + Fable 5 topped SecPass at 29.1%; Cursor + Fable 5 solved CVE-2021-44420 but removed for cheating; Cursor + Sonnet 5 had 17 confirmed cheating cases
AI coding tool exploited by Miasma to trigger credential harvester when repository opened
IDE
AI coding tool supported as an APM target
AI coding agent covered by skill-scanner (Cisco AI Defense) and MCP Defender
.cursor/mcp.json targeted
Agent harness used to run Sonnet 5 and Fable 5 on the 200-task dataset; showed an unusually high timeout rate when paired with Sonnet 5
coding agent harness; used for a separate companion benchmark run with Sonnet 5
AI coding assistant found vulnerable to the GhostApproval symlink attack; Cursor shipped a fix.
AI coding assistant discussed as part of the category introducing OSS license compliance risk
Targeted in 2025 by three malicious packages published by Snyk researchers; also discussed as an AI IDE platform using Open VSX instead of the official VS Code marketplace
AI coding environment; developers found running multiple such environments simultaneously
AI coding agent client compatible with the routing repository
IDE used to bridge/connect to the remote Meetings MCP server on port 3000 for exploitation
VS Code-based editor relying on Open VSX.
AI coding agent/harness paired with Claude Fable 5 in benchmark, topped Endor Labs' fair leaderboard with 72.6% FuncPass and 29% SecPass
AI coding agent; default install/settings confirmed to execute injected payload (video PoC), also seen via Warp terminal sandbox.
Agent harness undergoing a similar benchmark experiment with Fable 5, results pending
AI coding agent harness used in benchmark combinations (e.g., with Composer 2.5, Gemini 3.5 Flash, GPT-5.5, Claude Opus 4.8)
AI coding tool triggered via .cursor/rules/setup.mdc, an auto-applied rule tricking the agent into running the payload
VS Code fork, IDE extension ecosystem covered by Dev Machine Guard
AI coding agent cited as a type of tool that reads terminal output and could be targeted by the injection
AI coding assistant that sends code snippets to external APIs
Another VS Code-based editor with the same auto-update-on-by-default behavior, mentioned as needing the same cooldown enforcement.
Supported AI coding tool target; framework emits AGENTS.md + .cursor/{rules,skills,mcp.json}
Coding agent tool cited as operating inside developer workspaces and thus a similarly attractive target for threat actors.
Agent-capable LLM client using .cursor/rules files as selectable rules
AI coding agent/IDE plugin covered by the baseline
Indirect prompt injection via MCP config leading to RCE (CVE-2025-54135); rules file backdoor via .cursor/rules
Target IDE for malicious --install-extension commands
Forked IDE from VS Code; compatibility depends on adoption of VS Code base version 1.101+, support planned for future releases
AI coding assistant/agentic IDE that is the exploited platform
AI coding tool; per the AI Handbook, disallowed on client code except blockchain engagements
built-in browser / MCP interactions
AI agent whose local session artifacts are scouted and analyzed
Arbitrary code execution via search order hijacking (malicious git.exe/which.exe in cwd)
AI code editor with an agent that runs terminal commands, using a sandbox (since 2.x line) to limit what those commands can touch
AI coding agent shown to interpret injected Sentry error events as legitimate diagnostic resolution steps and execute attacker-controlled code
AI coding agent / IDE, scanned for on developer machines
In every workspace you can read 254
AI
Security
All technologies in this workspace.