Repository navigation
Conversation
|
@codex review |
|
The latest updates on your projects. Learn more about Vercel for GitHub. 1 Skipped Deployment
|
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: a178cde031
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".
| const body = | ||
| m.message.length === 0 && m.fanout | ||
| ? serializeFanoutComposite(m.fanout, (id) => id) | ||
| : m.message; |
There was a problem hiding this comment.
When a user reports a response produced from selected note text, a web selection, or an image, that input is stored in message.context/message.content and sent by buildPromptBlocks, while this formatter serializes only m.message. The resulting chat-transcript.md therefore omits the contextual input that triggered the problem, despite the source description promising that shared note text is included; serialize the relevant context/content or retain the effective sent prompt for reporting.
Useful? React with 👍 / 👎.
| if (selected.has("chatTranscript")) { | ||
| logs.push({ | ||
| id: "chatTranscript", | ||
| name: CHAT_TRANSCRIPT_NAME, | ||
| text: this.params.chatTranscript, | ||
| }); |
There was a problem hiding this comment.
When a long-running chat's transcript is larger than the bundle room left after the screenshot and Copilot log, buildReportBundle truncates this source to consume essentially all remaining bytes because sources are admitted in request order. Since the unbounded transcript is placed before the OpenCode and activity logs, those later sources are then dropped even though the nearby ordering policy exists specifically to prevent one large source from starving the others; put the bounded diagnostic logs first or reserve capacity for subsequent sources.
Useful? React with 👍 / 👎.
| defaultChecked: true, | ||
| } | ||
| ); | ||
| if (this.params.chatTranscript != null) { |
There was a problem hiding this comment.
This newly added conditional behavior has no nearby explanation containing the originating GitHub issue URL, and its corresponding buildSources() test description also omits that URL. Add the issue-linked why comment and include the same full URL in the it(...) description so this edge path retains its required provenance.
AGENTS.md reference: AGENTS.md:L28-L28
Useful? React with 👍 / 👎.
| origin: rootEl ?? undefined, | ||
| activeBackend: evidence?.backendId ?? settings.agentMode.activeBackend, | ||
| initialNote: evidence ? formatChatReportNote(evidence) : undefined, | ||
| chatTranscript: formatAgentChatTranscript(messages), |
There was a problem hiding this comment.
The manual More Actions → Report error path is changed here, but no test renders AgentHome and verifies that the active chat's transcript reaches openReportIssueModal; the existing controls test stops after invoking a supplied callback, while the new AgentChatMessages assertion covers only the separate feedback-card path. Consequently this line can regress or be removed while all added tests remain green, so cover the observable manual-report flow with the active session's messages.
AGENTS.md reference: AGENTS.md:L29-L29
Useful? React with 👍 / 👎.
a178cde to
fa441a8
Compare
46c1833 to
f64e6d3
Compare
fa441a8 to
51f7fe4
Compare
f64e6d3 to
7c2c3bc
Compare
51f7fe4 to
4652975
Compare
7c2c3bc to
846cace
Compare
4652975 to
ede8d19
Compare
The feedback card's Review & report now offers This chat's messages as a ticked attachment, written in the saved chat note's format and cleaned like every log. Settings' Report an issue has no chat and does not offer it. Relates to Brevilabs/obsidian-copilot-private#672 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ede8d19 to
96f1b2d
Compare
zeroliu
changed the title
|
Dropped: the chat's messages stay out of reports; #3616 keeps Report error without them. |
Relates to Brevilabs/obsidian-copilot-private#672
Why
A report started from the feedback card's Review & report is about that conversation. Today the bundle carries a screenshot of the window, which shows only the part of the chat on screen, plus logs. The Agent Mode activity log (on by default) holds the conversation only as raw protocol frames, with replies split into streamed chunks, and the bundle keeps only its tail, so early turns of a long chat are cut. When a user reports "it replied in Swedish", we need the turn where that happened, not just the last screenful.
What
Settings → Report an issue has no chat behind it and does not offer the item. Unticking it leaves the transcript out entirely.
Non goal
Screenshot
No screenshot yet for the feedback card path; the dialog's new This chat's messages item is covered by the
ReportIssueFlowstory.Risk
High
ReportIssueModaltests cover offering the item only from a chat, packingchat-transcript.mdinto a real zip, and its place in log order;formatAgentChatTranscriptandAgentChatMessagestests cover the transcript text and the feedback card passing itReview: read
buildSourcesandprepareinsrc/agentMode/ui/ReportIssueModal.tsx, the feedback card's handler insrc/agentMode/ui/AgentChatMessages.tsx, andformatAgentChatTranscriptinsrc/agentMode/session/AgentChatPersistenceManager.ts, then run Verification steps 2-5, including the redaction check in step 5.Verification
chat-transcript.md. Show zip and open it; the file holds the whole conversation as**user**:/**ai**:blocks with timestamps. Press Cancel.chat-transcript.mdis absent from the review page.sk-proj-abcdefghijklmnopqrstuvwxyz0123456789, then get the feedback card as in step 2, press Review & report, prepare the report, and openchat-transcript.mdfrom Show zip. The key is replaced by a visible redaction marker. Press Cancel.