Security advisories, builder stories, and engineering notes on AI agents, MCP, and the open standards behind them.
210 articlesRSS
The best Kimi plugins in 2026: 10 picks for Kimi Code and the Kimi app, with what each manifest loads, what you need before it works, HOL scan findings, verified install commands, example prompts, and pros and cons.
HOL GuardQuietOS is a runtime governor for AI agents, with tiered action decisions and a phone-based approval bridge.
An operator checklist for AI agent tool calls: inspect arguments, accounts, data, approval scope and the actual result before trusting an action.
HOLTest AI agent permission boundaries with synthetic files, mock services and a six-case worksheet. Separate model refusals from enforced restrictions.
HOLA practical guide to least privilege for AI agents: limit identities, files, tools and access duration, with a worked support-ticket example.
HOLThe best Cursor plugins in 2026: 10 picks with what each one loads into Cursor (rules, skills, agents, commands, hooks, MCP), HOL scan findings, verified install commands, example prompts, and pros and cons.
HOL GuardSix MCP tools for B2B prospecting with b2bsearch Actors on Apify, including configurable per-call spend caps.
A guide to the Seminara MCP extension and how its tool rules work with HOL Guard's review policies.
CoMPaSS is an unbenchmarked prompt framework that starts with user intent. Its creator explains the method and a plan to evaluate it.
Ten of the most-starred Claude Code plugins, one per job, with the HOL scan findings behind each label and install commands from each README.
HOL GuardGareth Hoyle found that anyone with a URL could download all 765 products from his AI-built Claude skills store for free. How it happened, four more problems the fix turned up, and the checks he now runs before shipping.
Gareth HoyleART Prompt Framework structures AI prompts around a role, request, and terms. Its creator explains the approach and when to use it.
Principal Engineer OS is an open-source agent skill for examining engineering decisions, failure modes, and recovery before shipping code.
Threat guides and evidence dossiers on prompt injection, MCP tool poisoning, slopsquatting, and the attacks shaping how teams ship code with agents.
How hidden instructions trick coding agents into unsafe work.
Read the researchEvergreen guideTool poisoning, overbroad permissions, and shadow MCP servers.
Read the researchNew dossierSlopsquatting is the supply-chain attack in which attackers register package names that AI models reliably hallucinate; when an agent writes an import for a package that never existed, the attacker’s registered code installs instead.
Read the researchNew dossierMalicious extensions and marketplace fraud place trojanized editor extensions, plugins, or CI actions inside trusted marketplaces, harvesting source code, prompts, and credentials from every developer who installs them.
Read the researchKeep exploring
The engineering desk continues across the ecosystem: security research on shipping agents safely, and plugins that put the standards to work.
