Problem
The Codex TUI currently uses the $ sigil for explicit skill and app mentions, for example -skill. Users who move between popular AI coding agents often expect @... mentions, so Codex's special $ convention adds avoidable friction.
Proposal
Add an opt-in TUI setting that keeps the current behavior by default but lets users choose @ for skill and app mentions:
[tui]
tool_mention_sigil = "at" # default: "dollar"
The setting should be applied end to end: config parsing and schema, composer insertion, mention popup/search behavior, submitted input parsing, and backend skill/app mention collection. Existing $ behavior should remain the default for compatibility.
Compatibility notes
- Default behavior remains
$.
- The
@ mode needs boundary checks so email addresses, scoped package names, and file/path fragments are not parsed as tool mentions.
- Plugin text mentions already have separate
@plugin behavior, so this should stay scoped to skill/app mentions unless maintainers want a broader plugin change.
Prototype
I have a local prototype branch that implements this behind the opt-in setting and adds targeted tests: https://github.com/alexandersumer/codex/tree/feat/configurable-tool-mention-sigil
I can open a PR if/when maintainers confirm this is a desired direction and the contribution policy allows it.
Problem
The Codex TUI currently uses the
$sigil for explicit skill and app mentions, for example-skill. Users who move between popular AI coding agents often expect@...mentions, so Codex's special$convention adds avoidable friction.Proposal
Add an opt-in TUI setting that keeps the current behavior by default but lets users choose
@for skill and app mentions:The setting should be applied end to end: config parsing and schema, composer insertion, mention popup/search behavior, submitted input parsing, and backend skill/app mention collection. Existing
$behavior should remain the default for compatibility.Compatibility notes
$.@mode needs boundary checks so email addresses, scoped package names, and file/path fragments are not parsed as tool mentions.@pluginbehavior, so this should stay scoped to skill/app mentions unless maintainers want a broader plugin change.Prototype
I have a local prototype branch that implements this behind the opt-in setting and adds targeted tests: https://github.com/alexandersumer/codex/tree/feat/configurable-tool-mention-sigil
I can open a PR if/when maintainers confirm this is a desired direction and the contribution policy allows it.