Repository navigation
Conversation
…, which hands your waiting comments to an agent The prompt says how many comments you left and on which drop, lists each with its anchor, words, replies and images, and ends with the Drop MCP calls that resolve them: read_drop, then a new version with resolves (accept_suggestion first when there are suggestions), or reply with resolve: true on a site or a board. It counts open threads where you spoke last, so a thread someone answered, or one an agent wrote for you, drops out.
Codex Review Summary: ❌ FailedThis 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. |
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
There was a problem hiding this comment.
Important
The copied prompt can ask agents to make changes Drop won't allow, and some feedback is missing from the handoff. Fix those cases before merging.
Approach: Reusing the viewer’s comments and Markdown formatter makes sense, but the generated actions need to match what the MCP tools can read, change, and resolve.
Reviewed changes — I traced the viewer’s copy action through thread selection, formatting, and the MCP handlers.
- Viewer action: The review pill offers Copy prompt when qualifying threads exist; the comments panel retains its all-feedback copy action.
- Thread selection: The last human message determines whether a thread is waiting on the current reader’s behalf, including follow-ups after resolution.
- Agent handoff: The prompt includes numbered feedback and drop-specific tool instructions. All 58 existing feedback and MCP tests passed; seven temporary integration checks reproduced the cases below.
gpt-6.2-sol(high) | 𝕏
Queued New pull request evidence is queued. |
There was a problem hiding this comment.
ℹ️ No P0 or P1 findings remain; one P2 issue with small screenshot annotations is noted inline.
Reviewed changes — Re-reviewed 6704816 against the previous review and ran the feedback, MCP tools, and site review suites. All 112 tests passed.
- Aligned thread selection with the stored open status used by MCP.
- Updated MCP guidance to match drop kind and reviewer permissions, and to publish again after accepting a suggestion only when edits remain.
- Added screenshot context to site prompts and made the copy action wait while screenshot metadata is loading.
gpt-6.2-sol(high) | 𝕏
| function siteAnchor(comment: DropComment, screenshot: Screenshot | null, url: string) { | ||
| const { label, ox, oy, ow, oh } = comment | ||
| const element = label ? `\n Element \`${label}\`` : "" | ||
| const box = `\n Box ${pct(ox)}–${pct(ox + ow)} across, ${pct(oy)}–${pct(oy + oh)} down the ${screenshot ? "screenshot" : "viewport"}` |
There was a problem hiding this comment.
P2 — Class: coordinate rounding collapses small annotation boxes. A site screenshot box with ox: 100, oy: 200, ow: 3, and oh: 4 copies as 10%–10% across, 20%–20% down, so an agent sees a point instead of the area shown in Drop. Preserve tenths of a percent in siteAnchor so the copied box matches the screenshot overlay.
…he pill, and Copy prompt steps aside while you comment, so the pill keeps one row

Once you've left a comment on a drop, the pill shows Copy prompt on one track with Comments and the eye, since all three act on the comments. It steps aside while you comment, so the comment actions keep the pill to one row. It copies a prompt you can paste into an agent:
Checked in headless Chromium on a local build: the button shows only once comments exist, and the click puts the text above on the clipboard. Unit tests cover which threads count and the wording per kind.
🤖 Generated with Claude Code