Beam / field notes

What we’re seeing in the field.

What we're seeing as we watch AI coding agents work — flagged actions, skill/MCP scan findings, and how to keep Claude Code, Codex, and the rest honest without slowing them down.Page 2 of 5

All authorsYash ThakkerBeam TeamNewest firstOldest firstTitle A–Z

AllAI AgentsSecuritySupply ChainAI Agent SecurityMCPAnthropicClaude CodeThreat IntelligencePrompt InjectionGovernanceLocal-firstCursorComplianceIncident TimelineAgent SkillsBeam

57 posts

Sep 14, 2026 · 6 min readAI agent deployment security and incident responseConnect deployment review to incident investigation: inspect container and cloud boundaries, understand telemetry gaps, and distinguish a suspicious instruction from a confirmed effect.Sep 14, 2026 · 7 min readAI agent security assessment: skills, tools, and evidenceA practical review of the path from untrusted instructions to sensitive actions, with a reproducible test design and clear limits for offline scanning.Sep 14, 2026 · 6 min readAI agent web and transport securityAgent security extends through the browser, session API, tool dispatcher, and provider connection. This guide maps those boundaries and shows how to review them without exposing real conversations or credentials.Sep 14, 2026 · 6 min readAI Asset Inventory and Model Security: A Repository ReviewA practical repository review for agents, model dependencies, prompts, datasets, and deployment configuration. Learn how to record evidence, prioritize findings, and distinguish configuration from runtime activity.Sep 14, 2026 · 6 min readUse Beam security skills with Claude Code and Codex sub-agentsA guide to choosing Beam security workflows, installing reviewer profiles, and combining specialist findings without losing scope or evidence.Sep 14, 2026 · 10 min readMCP security: review servers before connecting your agentA practical pre-connect review for local and remote MCP servers, covering package provenance, untrusted tool manifests, OAuth boundaries, and the limits of static scanning.Sep 14, 2026 · 10 min readHow to Stop Claude Code From Running Dangerous CommandsClaude Code will run a destructive shell command unless something stops it first. This guide walks through the three real levers — permission rules, PreToolUse hooks, and sandboxing — and where Beam fits as a local observer on top of them.Sep 14, 2026 · 10 min readHow to Stop Codex CLI From Running Dangerous CommandsOpenAI's Codex CLI sandboxes and gates commands by default, but the wrong approval/sandbox combination — or a plain --dangerously-bypass-approvals-and-sandbox flag — removes those guardrails entirely. Here's how the approval policy and sandbox actually work, and how to set safe defaults for a real repo.Sep 14, 2026 · 11 min readHow to Stop Cursor From Running Dangerous CommandsCursor's Agent mode can run real shell commands against your machine. This guide covers Run Modes, the terminalAllowlist in permissions.json, why the old command denylist was deprecated, and how to scope auto-run safely.Sep 14, 2026 · 11 min readHow to Stop Grok From Running Dangerous CommandsxAI's Grok Build CLI ships a permission and sandbox system of its own, but most developers reach Grok through a third-party harness where the model has no say over what runs. Here's how to lock down both cases.Sep 14, 2026 · 13 min readWhat are agent harnesses? And how to stop destructive commandsThe model only ever produces text — it is the harness around it that opens files and runs shell commands. This post defines "agent harness," maps the recurring safety mechanisms across Claude Code, Codex, Cursor, and Grok Build, and explains why observability is a second layer, not a replacement for sandboxing.Sep 14, 2026 · 7 min readWhy Beam is open source: agent safety belongs to everyoneEvery core Beam capability — the local collector, the scan heuristics, the agent hooks, and the security skills collection — is AGPL-3.0 open source you can read and run today. This explains what is open, what is licensed, and the reasoning behind the split.

Page 2 of 5 · 57 posts