Security
SkillFortify
Static analysis, provenance checks and a bill of materials for agent skills, so a skill is judged before it runs with your agent's privileges.
CodeElastic-2.0
Topic hub · Security
Check a skill before it runs with your agent's privileges.
An agent skill is code that runs with the agent's own authority, often installed from a registry with no review. Skills need a threat model, static analysis, provenance checks and a bill of materials.
Start here
Read it in this order.
Your progress is saved in this browser.
Reading path · 0 of 5 read
- PaperFormal Analysis and Supply Chain Security for Agentic AI Skills
- ToolSkillFortify: Agent Skill Security Analysis
- PostPHP2Shell: Patched PHP Can Still Be Vulnerable Until You Restart It
- PostAgent Skills Are Now Production Code. Most of Them Are Unverified.
- PostTwo-Thirds of Executives Already Leaked Data Through AI Agents. Here's What Engineers Can Actually Do About It.
Tools
What to run.
Research
The papers.
Formal Analysis and Supply Chain Security for Agentic AI Skills
96.15% F1, five theorems with full proofs, and a published negative result on information-flow analysis
Glossary
Terms you'll meet.
- Agent skill
A packaged capability an agent can install and run, executed with the agent's own authority.
- Skill supply chain
The path a skill takes from author to registry to your agent. A registry with no review gate makes every install a trust decision.
- Skill bill of materials
An inventory of what a skill contains and depends on, generated so it can be reviewed before the skill executes.
Writing
Posts on security.
Field reports
PHP2Shell: Patched PHP Can Still Be Vulnerable Until You Restart It
You installed the PHP2Shell fix and your site can still answer as vulnerable. Where PHP runs in the request path, how to prove which build every instance is serving, reload versus restart per SAPI, the third issue with no fix release yet, and the defender loop.
18 min read
Field reports
Agent Skills Are Now Production Code. Most of Them Are Unverified.
Nine major agent announcements in three weeks. When agents can refactor codebases, touch ERPs, and control robots, verification at the skill level is no longer optional. Here's exactly what developers need to do.
5 min read
What's changing
Two-Thirds of Executives Already Leaked Data Through AI Agents. Here's What Engineers Can Actually Do About It.
The math is brutal — a 32-step agent at 95% per-step accuracy yields 19% end-to-end success. Five open-source tools that fix AI agent reliability.
6 min read
Field reports
AI Agents Need an Iron Dome Before They Get an Iron Man
341 malicious skills. 135K GitHub stars. 1.5 million leaked API tokens. The OpenClaw crisis proved what we've been saying: AI agent security isn't a feature — it's an existential requirement.
7 min read
Field reports
Your AI Agent Has Root Access. Its Skills Don't Get Checked.
Your AI coding agent can read every file on your machine. It can write to any directory. Execute...
4 min read
Field reports
392 Skills. Zero Verification. That Is the State of AI Agent Security.
Claude Code has 392 skills. Cursor has plugins. Every agent framework has extensions. GitHub Copilot...
3 min read
Field reports
Operation Pale Fire: What Block's Red Team Proved About AI Agent Security
Block's security team ran a red team exercise against their own AI agent Goose and achieved full compromise. The findings reveal architectural vulnerabilities that affect every AI agent connecting to external tools via MCP.
6 min read
Field reports
The 5 Security Risks Nobody Talks About in AI Coding Agents
In January 2026, Block's security team ran a red team exercise against their own AI agent, Goose....
10 min read