Awesome list for AI agent harness engineering: tools, patterns, evals, memory, MCP, permissions, observability, and orchestration.
Python agent-harnessagent-memoryagent-orchestrationai-agent-harnessai-agentsawesome-listcontext-engineeringharness-engineeringmcp
Add 'What's in Your Agent's Context?' context privilege escalation paper (arXiv 2609.01222) to Security, Sandbox & Permissions
ai-boostfad270e230da
diff --git a/README.md b/README.mdindex e455573..cb7b149 100644--- a/README.md+++ b/README.md@@ -543,6 +543,7 @@ Real repositories worth studying — each with a note on *why* it's worth your t- [Prompt Injection — Simon Willison's Series](https://simonwillison.net/series/prompt-injection/) — The most thorough public writing on why indirect prompt injection is uniquely dangerous for agent harnesses: agents actively consume untrusted external content (emails, web pages, tool outputs) that can hijack their actions. Essential for understanding the attack surface before designing trust boundaries.- [OWASP LLM01:2025 — Prompt Injection](https://genai.owasp.org/llmrisk/llm01-prompt-injection/) — OWASP's authoritative classification of direct and indirect prompt injection risks. Complements the tldrsec defense catalog: use this to define the threat model, use tldrsec to select countermeasures.- [StackOne Defender](https://github.com/stackoneHQ/defender) — Open-source indirect prompt injection defense for agents: 22MB CPU-only model, ~4ms latency, 89% balanced accuracy. Inspects tool results before they enter the LLM context window, turning untrusted MCP/CLI/function-call output into a harness-sanitized boundary rather than a prompt-level gamble. +- [What's in Your Agent's Context? Context Privilege Escalation Attacks against AI Agent Harness](https://arxiv.org/abs/2609.01222) — September 2026 first systematic security analysis of how harnesses *assemble* context across 12 real-world systems (including Claude Code and Codex), introducing two attack classes: MessageRole Context Privilege Escalation (attacker-controlled content promoted into higher-privileged message roles) and Cross-Scope CPE (attacker content persisting beyond its original scope), with consequences up to full agent compromise and RCE. Essential because prompt-injection defenses assume a static trust model — this paper shows the context-assembly layer itself is a privilege-escalation surface that harness design must treat as a first-class boundary.- [Daytona](https://github.com/daytonaio/daytona) — OCI-container sandboxes with sub-90ms startup, built-in Git operations, LSP support, and indefinite state persistence. Complements E2B for harnesses that need long-lived working directories across multiple agent sessions rather than ephemeral code execution. - [NeMo Guardrails](https://github.com/NVIDIA-NeMo/Guardrails) — NVIDIA's programmable guardrails toolkit: define input, dialog, retrieval, execution, and output rails that intercept the agent loop at five distinct layers using the Colang DSL. The execution rail layer specifically governs what tools the LLM can invoke and what their inputs/outputs may contain — the reference for behavioral-level enforcement when static allow/deny lists are insufficient. - [LangSmith Sandboxes: Secure Code Execution for Agents](https://blog.langchain.com/introducing-langsmith-sandboxes-secure-code-execution-for-agents/) — Describes a microVM-based sandboxing architecture with kernel-level isolation, resource caps (CPU/memory/disk), and an authentication proxy that keeps secrets entirely out of the runtime environment. Persistent WebSocket sessions support long-running agent tasks like dependency installation and test suite execution without the overhead of per-call container restarts.