Agent Profile selector

Resolved 💬 3 comments Opened Dec 30, 2025 by QIANSUIMINGMINGMING Closed Feb 23, 2026
💡 Likely answer: A maintainer (etraut-openai, contributor) responded on this thread — see the highlighted reply below.

Hi Codex team,

Problem

In real workflows, I need to switch instruction sets frequently for different general tasks (e.g., coding vs. paper-writing vs. research/planning), each of which triggers different agents.md and skill sets. Meanwhile, some monorepo contains multiple sub-tasks (e.g., preliminary experiment planning & complex system implementation).
Today, the extension pulls instructions via the CLI discovery mechanism (global/project AGENTS.md), which makes switching cumbersome and error-prone:

  • It requires renaming/moving ~/.codex/AGENTS.md, or
  • launching the IDE with different CODEX_HOME, or
  • symlinks/overrides that are fragile and hard to reason about.
  • manually copy the same agents.md for each monorepo

This is a UX mismatch: the extension already provides a convenient model switcher, but there is no equivalent for instruction profiles.

Proposed UX

Add an “Agent Profile” dropdown in the extension UI:

  • Shows current profile name
  • Lets users switch profiles in 1–2 clicks
  • Clearly indicates the active instruction source (Global / Workspace / Session)

Why this matters

Agent instructions are as impactful as model choice. A fast, explicit selector reduces mistakes and makes the extension usable across mixed workflows.

Thanks!

Additional information

_No response_

View original on GitHub ↗

3 Comments

glebsa8 · 8 months ago

Really missing that. Similar feature request:

Hope it will be implemented by the team. The community can't send PRs for that.

etraut-openai contributor · 8 months ago

You may find the new "skills" feature to be useful here. You can define different skills and invoke them explicitly by name using $skill_1 or $skill_2. Each skill can provide different instructions. For details, refer to this documentation.

Another option is to define custom prompts, which can be invoked using slash commands.

etraut-openai contributor · 6 months ago

This feature request hasn't received enojgh upvotes, so I'm going to close it.