Codex Unified @ Mention: Search Files, Plugins, and Skills with a Single Command
📖 Glossary
AI Box (also known as Agent Computer / Agent PC), is a dedicated local hardware device that runs AI Agents. Pre-installed with an AI agent management system, plug-and-play, running 24/7. Users can remotely command AI to work via Discord, Slack, Telegram, WhatsApp, and more.
Abstract: Finding files, calling plugins, and activating Skills in Codex previously required memorizing three different command sets. The new unified @ mention system lets you search files with @filename, call plugins with @pluginname, and activate Skills with @skillname. One symbol for all resource access — significantly improving development efficiency.
Anyone who builds with AI coding tools knows the feeling: tools get more powerful, but remembering how to invoke all their features becomes a burden of its own.
Codex's old problem: file search used one command syntax, plugin invocation used a different grammar, and Skill activation had yet another format. Developers constantly context-switched, losing efficiency to unnecessary cognitive load.
In a mid-June update, Codex introduced a unified @ mention system. All resources — files, plugins, Skills — can now be invoked with a single @ symbol.
How Unified @ Mention Works
@file: Type @ followed by a filename in conversation. Codex auto-searches the project for matching files. Supports fuzzy matching and path auto-completion. For example, typing @config lists all files containing "config." Select one and it's referenced directly in the current conversation context.
@plugin: Type @ followed by a plugin name to invoke installed plugin functionality directly. No separate plugin invocation commands needed. For example, @linter runs lint checking on current code, @formatter directly formats the current file.
@Skills: Type @ followed by a skill name to activate the corresponding Agent skill. For example, @review triggers code review skill, @test triggers automated test generation.
Why This Change Matters
On the surface it's just syntax unification. But it solves a deeper problem: cognitive switching cost.
Previous workflow: writing code → want to lint → switch to terminal → run lint command → switch back to editor → see results → continue coding. Multiple context switches.
Now: directly type @linter in Codex conversation → results return immediately → continue coding. The whole process stays within the current workspace.
For developers unfamiliar with command lines, this change lowers the entry barrier. No need to remember "which command for lint, which flag for format" — just @ it.
Difference from IDE Auto-Complete
Someone might ask: isn't this just IDE auto-complete?
The difference: @ mention operates at the semantic level, not the text level.
IDE auto-complete matches file paths and symbol names textually. @ mention is semantic — it understands what you want to do. Typing @test, Codex knows you want to generate test cases for the current function. It doesn't just open a test file — it analyzes the current function's logic and generates test code covering edge cases.
Typing @review, Codex doesn't simply hand you a code review checklist. It reads current code changes, identifies potential issues, and provides specific improvement suggestions with reasoning.
This is an upgrade from "tool invocation" to "intent understanding."
Implications for the Agent Ecosystem
Unified @ mention reflects the standardization of Agent capabilities. Files, plugins, and Skills are essentially three resource types Agents can invoke. Unifying them under one syntax means Agent resource scheduling has reached a new level.
For Kaihe AIBOX users, this trend is worth noting. Kaihe AIBOX uses a local multi-Agent + cloud LLM architecture, with multiple Agents running local tasks. When Agents need to invoke different resources — read files, call tools, use Skills — a unified scheduling interface can significantly reduce coordination overhead between Agents.
Codex's @ mention is unification within the editor. Kaihe AIBOX's Agent scheduling is unification at the system level. Same direction: converge distributed resource invocation through a single entry point, reducing complexity.
Real-World Experience
Developer community feedback:
"Before, with Codex, file references needed full paths, plugin invocation required memorizing command names, and Skills required reading documentation on how to activate. Now it's all @. This one change alone saves dozens of context switches per day."
"The most practical feature is fuzzy matching. Can't remember the complete filename? No problem — type @user and it lists all user-related files. Pick one. Way faster than navigating the file tree."
Data Sources
This article references OpenAI Codex changelogs, GitHub community discussions, and CSDN technical coverage.
-#KaiheAIBOX #AIAgent #OpenSource #ArtificialIntelligence
Kaihe AIBOX | The Agent Computer That Works 7×24 for You · AI Agent