[codex] Add hybrid skill discovery and composer UX#36
Conversation
- Discover project and user skills for Codex and Claude - Allow composer skill tokens to flow through orchestration and turns - Make auth migrations self-healing on existing local databases
- Increase command description truncation on the server - Add hover tooltips for skill entries in the composer menu
github-actions
bot
added
size:XXL
- Derive and dedupe skill selections from discovered Codex commands - Keep composer cursor and mention handling aligned with skill tokens - Normalize workspace paths before listing provider commands over WS
- Replace skill-specific composer chips with generic custom tokens - Move Codex skill menu/selection logic into the chat skill extension
- Move mention and custom token Lexical nodes into a shared module - Keep ComposerPromptEditor focused on composer behavior
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d8a0bfba5b
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| promptRef.current = next.text; | ||
| setPrompt(next.text); | ||
| setComposerCursor(nextCursor); |
There was a problem hiding this comment.
When a pending user-input question is active, composer edits are supposed to flow through onChangeActivePendingUserInputCustomAnswer (as done in applyPromptReplacement) so the response payload matches what the user sees. The skill-selection branch bypasses that path and writes directly via setPrompt, so picking a $skill suggestion during pending-question mode can update the editor text but not the pending answer state that is ultimately submitted.
Useful? React with 👍 / 👎.
1 participant
O que mudou
~/.codex/skills/.systeme também caminhos extras em~/.agents/skillsturn/start, em vez de depender só de texto no promptPor que mudou
O fluxo anterior tinha três problemas reais:
Impacto
Validação
bun fmtbun lintbun typecheckObservações
bun lintcontinua com um warning preexistente emapps/web/src/environments/runtime/catalog.test.ts