Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

Greptile Summary

This PR syncs the address-review slash command with upstream improvements, adding multi-page API pagination support (jq -s), node_id fields for thread resolution, a GraphQL query to fetch and filter already-resolved review threads, and several UX clarifications (single sequential numbering across triage categories, explicit DISCUSS item handling, correct reply path for review summary bodies).

Key changes:

• --paginate + jq -s '[.[].[] | ...]' pattern correctly aggregates multi-page REST responses
• GraphQL query with --paginate correctly uses a nullable $endCursor:String variable and pageInfo { hasNextPage endCursor } for outer thread pagination
• -F pr={PR_NUMBER} (capital -F) correctly passes the PR number as an integer matching the $pr:Int! GraphQL type
• Removal of the incorrect "standalone review comment" POST /pulls/{PR_NUMBER}/comments path in favor of the /replies endpoint (which works for all existing comments) is a correctness improvement
• The activeForm field was removed from Step 5 task-creation instructions; since TodoWrite requires this field, restoring explicit guidance would reduce the risk of a malformed tool call
• The known comments(first:100) inner-pagination limitation is now documented in Known Limitations

Confidence Score: 4/5

• Safe to merge — all shell commands and jq filters are correct; one minor documentation gap around activeForm in Step 5.
• The pagination fix (jq -s), GraphQL thread-resolution query, and resolved-thread filtering are all technically sound. The only concern is the removal of activeForm guidance from the TodoWrite instructions, which could lead to malformed tool calls in edge cases but is unlikely to cause failures in practice given Claude's tool-schema awareness.
• No files require special attention beyond the activeForm note in .claude/commands/address-review.md.

Important Files Changed

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A(["/address-review input"]) --> B["Step 1: Detect repo\ngh repo view"]
    B --> C["Step 2: Parse input\nPR number / URL / review ID"]
    C --> D{"Input type?"}
    D -->|issuecomment-ID| E["GET /issues/comments/ID"]
    D -->|pullrequestreview-ID| F["GET /pulls/PR/reviews/ID\nGET /pulls/PR/reviews/ID/comments\n--paginate + jq -s"]
    D -->|PR number only| G["GET /pulls/PR/comments\n--paginate + jq -s\nGET /issues/PR/comments\n--paginate + jq -s"]
    F --> H
    G --> H["GraphQL: fetch reviewThreads\nwith node_id + isResolved\n--paginate + jq -s"]
    H --> I["Filter:\n- Skip resolved threads (NEW)\n- Skip in_reply_to_id replies\n- Deduplicate bots"]
    I --> J["Step 4: Triage\nMUST-FIX / DISCUSS / SKIPPED"]
    J --> K["Step 5: TodoWrite\nMUST-FIX items only"]
    K --> L["Step 6: Present triage\nSingle sequential numbering (NEW)\nWait for user selection"]
    L --> M{"User selects items"}
    M -->|"MUST-FIX or DISCUSS"| N["Address in code/tests/docs"]
    M -->|"SKIPPED rationale"| O["Post rationale reply only"]
    N --> P{"Comment type?"}
    P -->|"Issue comment"| Q["POST /issues/PR/comments"]
    P -->|"Review comment"| R["POST /pulls/PR/comments/ID/replies"]
    P -->|"Review summary body (NEW)"| S["POST /issues/PR/comments\nas general PR comment"]
    Q --> T{"Thread resolvable?"}
    R --> T
    S --> T
    T -->|"Yes + thread_id available"| U["GraphQL: resolveReviewThread"]
    T -->|No| V([Done])
    U --> V

_{Last reviewed commit: "Sync address-review ..."}

Overall this is a solid improvement. See follow-up comments for specifics.

Issue 1: node_id missing from single-comment fetch (Step 3, issuecomment path)

The issuecomment fetch path omits node_id, while every other comment-fetching path in Step 3 now includes it. Issue comments do not belong to review threads so this will not break thread resolution in practice, but the inconsistency is confusing. Suggest adding node_id: .node_id here for uniformity.

Issue 2: Removal of standalone review comment creation path

The old section for standalone review comments documented how to create a new anchored comment on a line (using commit_id, path, line, side). This is removed in favour of using the /replies endpoint for all existing review comments. Using /replies for existing threads is correct, but if a thread ID is unavailable or a comment has been deleted there is no longer a documented fallback. Worth a short note in Known Limitations suggesting the general PR comment endpoint as a fallback.

Issue 3: GraphQL inner pagination note is slightly incomplete. The Known Limitations note says comments(first:100) is hardcoded and may truncate threads with more than 100 comments. The reviewThreads(first:100) per page is also hardcoded. Very rare in practice, but both limits are worth mentioning. Minor nit: the jq -s slurp pattern for aggregating paginated REST results is a correct fix. A short inline comment explaining why jq -s is needed would help future readers (--paginate emits one JSON array per page; -s slurps them so .[].[] can flatten across pages). Overall: the core logic (GraphQL thread metadata, is_resolved filtering, ARGUMENTS variable wiring) is correct and well thought out. The databaseId/node_id mapping between GraphQL and REST is handled properly. See the separate inline comment for a readability suggestion on the long GraphQL query.