Summary

Enables the A2A comment-ID hand-off pattern across three coordinated phases. A reviewer captures commentId from manage_issue_comment's enriched response and mailboxes it to the peer; peer fetches just-this-comment via get_conversation({comment_id}) instead of re-walking the whole thread. Scales linearly with new-comment volume across multi-cycle reviews rather than cumulative thread size.

Empirical motivation from this session: by cycle 3 of #10269's three-cycle cross-family review, get_conversation was returning ~15K tokens of prior-round content that the current-cycle reviewer had already seen. This PR closes that context-waste path without introducing new tools (stays within the 84/100 tool-surface discipline — same condensation pattern as manage_database replacing start_database + stop_database).

Phases (all in one PR — three bites unblock each other)

Phase 1 — manage_issue_comment enriched response

• ADD_COMMENT GraphQL mutation extended with createdAt on the comment node.
• createComment now captures the response and returns {message, commentId, url, createdAt} — matching updateComment's existing {message, commentId, url, updatedAt} shape.
• openapi.yaml introduces a new CommentResponse schema extending SuccessResponse. Backward-compatible: existing consumers reading only message continue to work; new fields are additive.

Phase 2 — get_conversation selectors

• GET_CONVERSATION extended with id field on comment nodes.
• getConversation accepts object options {pr_number, comment_id?, since_comment_id?, last_n?}. Legacy positional prNumber form retained via type-discrimination at the entry point (backward-compat migration path).
• Selector precedence: comment_id > since_comment_id > last_n > full. Client-side filter post-fetch (single GraphQL query already caps at maxCommentsPerPullRequest).
• openapi.yaml switches get_conversation to x-pass-as-object: true with the new optional params and a structured 200 response schema.

Phase 3 — A2A propagation convention codified

• pr-review-guide.md §9 — A2A Comment-ID Hand-off Protocol (reviewer side). Workflow, selector precedence, anti-patterns.
• pull-request-workflow.md §8.1 — symmetric Author-side propagation for multi-cycle review response loops.

Test Evidence

npx playwright test -c test/playwright/playwright.config.unit.mjs \
  test/playwright/unit/ai/mcp/server/github-workflow/IssueService.spec.mjs \
  test/playwright/unit/ai/mcp/server/github-workflow/PullRequestService.spec.mjs

• IssueService.spec.mjs (8 tests): create-action enriched-return assertion (issue + PR paths), update-action regression lock, input validation guards, GraphQL error propagation.
• PullRequestService.spec.mjs (13 tests): default full-fetch, each selector in isolation, selector precedence when multiple passed, invalid-id behavior (empty result, no silent fallback to full), legacy positional form, structured errors.

Stepping Back — Reflection Protocol

Considered and explicitly rejected:

• Splitting into three PRs (one per phase). Phase 3 docs reference Phases 1 and 2 mechanics; a split would land aspirational-convention docs before the substrate they describe. Bundling demonstrates the complete pattern in one reviewable artifact.
• Server-side pagination via GraphQL cursor for selectors. Considered, rejected as premature. Current PR conversation sizes (dozens of comments at most) make client-side filter adequate. If empirical volume demands server-side, it's a follow-up — doesn't block the A2A pattern.
• Extending manage_discussion_comment symmetrically in this PR. Deferred as out-of-scope. The pattern translates directly — same {commentId, url, createdAt} enrichment applies — but keeping this PR focused on the ticket's stated manage_issue_comment scope. Natural follow-up ticket if cross-Discussion A2A propagation warrants.

Out of Scope

• Discussion-comment parallel enrichment (manage_discussion_comment / DiscussionService.addCommentToDiscussion). Deferred; pattern translates directly.
• Server-side pagination via GraphQL after cursor for since_comment_id. Client-side filter works for current volumes.
• Retroactive tooling for existing long threads. Forward-looking only.

Related

• Origin: 2026-04-24 session empirical motivation — #10269 three-cycle review context-fetch growth.
• Cross-family endorsement of the shape: @neo-gemini-pro's 2026-04-24 01:06 reply confirming the comment-ID hand-off value.
• Related meta-enhancement tickets from the same session: #10270 (ChromaManager boot warning), #10273 / PR #10279 (decile anchor rubric), #10274 / PR #10277 (merge-auth), #10275 (silence detection daemon), #10276 (ideation adjacency sweep), #10278 / PR #10282 (ideation iterative workflow), #10284 (MailboxService post-linkNodes verification).
• Tool-surface condensation precedent: manage_database replacing start_database + stop_database.

Resolves #10272