- Category
- Cybersecurity
- Rank
- No. 1150Tools index
- Pricing
- Open Source
- Platform
- cli
- Type
- TOOL
- GitHub
- 8 stars
- Latest release
- v0.10.1
- Date
About
Stroq is a local, deterministic firewall that hooks into AI coding agents (Claude Code, Cursor, Codex, Copilot CLI, Windsurf, OpenClaw, MCP clients) to scan content the agent reads for injected instructions, taint the session, and block or ask before risky follow-up actions like shell execution, secret exfiltration, or external git pushes. It ships a built-in replay suite that tests the default policy against 13 real-world agent-hijacking incidents, all fully offline with no cloud dependency.
What it can do
Scan content read by AI coding agents for injected instructions
Content read by AI agent (files, web pages, tool output) → Detection of prompt injection attempts
Taint session state when malicious content is detected
Scan results → Tainted session flag
Block or prompt for confirmation before risky follow-up actions (shell execution, secret exfiltration, external git pushes)
Agent action request → Allow/block/ask decision
Run a built-in replay suite testing default policy against 13 documented real-world agent-hijacking incidents
Recorded attack scenarios → Pass/fail results per incident
Why it made the leaderboard
Blocks the pattern where an agent reads malicious instructions from a file or page and then acts on them, by tracking taint across the session and gating risky follow-up commands.
Tags
Tech Stack
Media
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.