Repository navigation
Studio: add managed project instructions, skills, and /init - #10576
PhilipJohnBasile wants to merge 6 commits into
Conversation
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![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: 87d3def71e
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".
|
This resolves project guidance server side and prepends it as a system message on the chat, Responses and Anthropic paths, so every project chat gets different context than today. Could you confirm what a project with no AGENTS.md sends, so existing chats are byte for byte unchanged, before I review? |
…xt-review-20260909
Add project /init, AGENTS.md discovery, and project skill guidance using Studio's existing managed workspace. The Instructions & skills panel shows resolved instructions, discovered skills and excluded paths. /init creates a starter file without overwriting existing guidance.
Projects with no AGENTS file or skills retain existing model messages and system bytes. Stored project instructions keep their existing frontend <project_instructions> formatting for sends and token counts. New guidance is added only when present. External requests always carry the project session independently of selected tools.
The request lease precedes project validation, and workspace revalidation cannot fall back to a deleted project. Current head
0394558ce98ba82c3ec3f73e91e4d6c4e686dba0includes upstream08733896d. Validation passed 83 backend cases, including guidance transport and generation-supervisor regressions, all 7,292 frontend tests, app/test typechecks, production build, and startup budget.Root and skill reads are bounded and reject symlinks. Windows keeps existing stored instructions but repository traversal and /init require the separate native filesystem layer. Nested instruction discovery during edits, folder selection, verification, Git operations, delegation and continuity are outside this split. See studio/PROJECT_GUIDANCE_REVIEW.md for scope and historical validation, which is distinguished from the current review pass.
Prompt compatibility confirmation (September 10): the transport regressions pass within the current 83-case backend run. With no AGENTS.md and no discovered skills, chat and Responses preserve the original model-message objects, and Anthropic preserves both string and structured system blocks, including whitespace and cache metadata. Existing stored project instructions keep their caller-provided serialization. A project with discovered skills intentionally receives the skill catalog even if it has no AGENTS.md; byte-for-byte preservation applies when neither source adds guidance.