PR Review Summary

Status: Comment (eligible for human merge)

Peer-Review Opening: Surgical and on-target — exactly the dispatch extension the ticket prescribes. The follow-up Cursor-removal commit upgrades the discipline from "1-line fix that adds speculative coverage" to "1-line fix with Truth-in-Code rigor." Two Required Actions: spec coverage and a stale-line in the PR body. The rest is polish.

🕸️ Context & Graph Linking

• Target Issue ID: Resolves #10467
• Related Graph Nodes:
  • Substrate predecessor: #10411 (PR that introduced the per-appName dispatch on 2026-04-27 and added Cursor's 'l' shortcut — speculatively, as it turns out)
  • Adjacent: #10446 (feat(bridge-daemon): fail-fast on empty appName metadata) — same file, complementary defensive guard
  • Empirical reproducer trail: this bug only surfaced once A2A came back from the SQLite IN-clause overflow arc (#10464 / #10466). Wake delivery wasn't routinely exercised while A2A was wedged.

🔬 Depth Floor

Challenge: Spec coverage gap. The existing test/playwright/unit/ai/scripts/bridge-daemon.spec.mjs (243 lines) has zero references to Antigravity, Cursor, or tabShortcut. The dispatch branch this PR modifies has never had spec coverage. This isn't introduced by this PR, but it IS the AC item #3 from #10467's edited body: "extended to cover the appName === 'Antigravity' branch (assert tabShortcut === 'l' is applied)." The empirical-no-corruption AC is satisfiable; the durable-test AC is not. Required Action below.

Pattern observation across the recent fix cohort (#10464, #10466, this PR): surgical fixes shipping without spec coverage, with spec deferred as "polish." Each individual deferral is reasonable; the cumulative drift is worth memorializing — see [RETROSPECTIVE] below.

Rhetorical-Drift Audit (per guide §7.4):

• PR description (commit-1 framing): "added appName === 'Antigravity' alongside Cursor to share the l shortcut behavior" — STALE relative to the latest diff. Commit-2 removed Cursor from the dispatch entirely; the PR body still describes the original combined-branch shape. Required Action below to update the body.
• Diff content matches the second commit's framing (fix: remove unverified cursor support) — Truth-in-Code accurate, no drift in code or its inline comments.
• No new Anchor & Echo summaries introduced beyond the Electron-Paradox comment update from "Antigravity, Cursor" → "Antigravity, VS Code." That comment update is materially correct: VS Code is the true bundle name for the Antigravity extension's host process; Cursor was an unverified addition.
• No [RETROSPECTIVE] tag in the PR description.
• Linked anchors (#10467) accurately establish the close-target.

Findings: Pass on code/comments. One drift in PR body — Required Action below.

🧠 Graph Ingestion Notes

• [KB_GAP]: None in the file's inline comments anymore — commit-2's comment update from "For Cursor, Cmd+L..." → "For Antigravity, Cmd+L..." closes the gap I had flagged in Cycle 1. Net positive over the original PR.
• [TOOLING_GAP]: No spec coverage on the per-appName dispatch in bridge-daemon.spec.mjs. Required Action below.
• [RETROSPECTIVE]: Two distinct lessons land in this PR:
  1. Truth-in-Code discipline. Commit-2's removal of Cursor mapping ("never explicitly tested for 2026, shouldn't give the impression of false support") is a sharper standard than "leave it in case it works." Worth memorializing as a swarm-level discipline: don't ship code paths claiming coverage you haven't verified. The cost of false-support is downstream agents trusting it and then debugging the failure path on a bug-hunt.
  2. Cohort pattern across the recent A2A-substrate-stabilization PRs (#10464, #10466, #10468) — surgical 1-line fixes shipping without spec coverage, AC items for spec deferred individually as "polish." Cumulative effect: three new code paths landed across two days with zero unit-test coverage on the per-appName dispatch + chunking branches. Not a blocker for any individual PR, but the pattern is worth a coordinated cleanup ticket so the spec coverage gap doesn't compound silently.

🛂 Provenance Audit

N/A — standard bug fix on an established dispatch substrate, not a major architectural abstraction. §7.3 threshold not triggered.

🎯 Close-Target Audit

• Close-targets identified: Resolves #10467
• For #10467: confirmed not epic-labeled (carries bug, ai, architecture, regression). Close-target valid per §5.2.

Findings: Pass.

📡 MCP-Tool-Description Budget Audit

N/A — PR does not touch any ai/mcp/server/*/openapi.yaml files.

🔌 Wire-Format Compatibility Audit

• Does the change impact downstream consumers? Removing appName === 'Cursor' from the dispatch is technically a contract narrowing. Impact: anyone running with appName: 'Cursor' configured in their harnessTargetMetadata will no longer get an automatic Cmd+L tabShortcut applied — they'd need to set tabShortcut: 'l' explicitly. If Cursor users existed (the empirical question @tobiu raised), they'd hit this regression. Given the speculative-support framing of commit-2, the assumption is no production Cursor users exist; that assumption is worth confirming with whichever swarm member added Cursor in #10411 if a swarm-instance audit hasn't already cleared it.
• No payload structure modified.
• Antigravity-side: backward compatible. Subscriptions that explicitly set tabShortcut: null (deliberate opt-out per the line-503 comment) are still respected since the dispatch only fires when tabShortcut === undefined.

Findings: Pass on the Antigravity path. Soft contract narrowing on the Cursor path — flagged for awareness, not blocking.

🔗 Cross-Skill Integration Audit

• No predecessor skill steps need updating — internal dispatch fix
• AGENTS_STARTUP.md §9 Workflow skills list does NOT need updating
• No reference files mention a predecessor pattern requiring update
• No new MCP tool surface added
• No new convention introduced — extends existing per-appName dispatch shape

Findings: All checks pass — no integration gaps.

📋 Required Actions

To proceed with merging, please address the following:

• Update PR body: the "Deltas from ticket" line currently says "added appName === 'Antigravity' alongside Cursor to share the l shortcut behavior, restoring the intended osascript flow for the Antigravity composer." That description matches commit-1, not the final state after commit-2 removed Cursor entirely. Tighten to: "replaced the speculative appName === 'Cursor' dispatch branch with appName === 'Antigravity' (commit-2 follow-up: removed unverified Cursor support per Truth-in-Code discipline)."
• Extend test/playwright/unit/ai/scripts/bridge-daemon.spec.mjs to cover the per-appName dispatch — assert tabShortcut === 'l' for appName === 'Antigravity', tabShortcut === '3' for appName === 'Claude', and tabShortcut === undefined for any unmatched appName. Closes AC item #3 from #10467 and retroactively fills the coverage gap that let this regression slip through #10411.

(Both polish-tier — author's call whether to land in this PR or fast follow-up. The empirical-no-corruption AC from #10467 is satisfiable today via npm run ai:run-sandman cycle into an Antigravity session.)

📊 Evaluation Metrics

• [ARCH_ALIGNMENT]: 100 — perfect alignment with the existing per-appName dispatch pattern, plus Truth-in-Code rigor in commit-2 (removing speculative Cursor coverage rather than retaining it). I actively considered code-style consistency (matches), pattern fit (uses same shape as Claude's branch), backward compatibility on the Antigravity path (deliberate-opt-out via tabShortcut: null still respected), and Truth-in-Code discipline on the Cursor path (commit-2 removes the false-support claim) and confirmed all align.
• [CONTENT_COMPLETENESS]: 80 — code-side comments are now accurate after commit-2 (KB_GAP from Cycle 1 closed). 20 points deducted because the PR body's "Deltas from ticket" line is stale relative to the final diff state — see Required Action above.
• [EXECUTION_QUALITY]: 70 — code change is correct and minimal. 30 points deducted because the existing bridge-daemon.spec.mjs has zero coverage of the per-appName dispatch (no references to Antigravity/Cursor/tabShortcut); AC item #3 from #10467 explicitly required this and is not delivered. The empirical-no-corruption verification is satisfiable but the regression-prevention test layer isn't.
• [PRODUCTIVITY]: 80 — directly resolves the wake-editor-corruption bug for every Antigravity-harness agent in the swarm, and improves the substrate's Truth-in-Code posture by removing speculative Cursor support. 20 points deducted because the AC item on spec coverage is deferred — same pattern as the recent #10464 / #10466 fix cohort (see [RETROSPECTIVE]).
• [IMPACT]: 50 — moderate: surgical fix on the substrate that affects all Antigravity-harness wake delivery. Concrete: every cross-family A2A from a non-Antigravity agent to a @neo-gemini-pro (or any future Antigravity-harnessed identity) was corrupting the editor pre-fix. Substantial substrate-stability impact, even though the diff is one line of dispatch + two-line comment polish.
• [COMPLEXITY]: 10 — Trivial: one-line dispatch change + two-line comment polish; no cross-substrate integration; no new abstractions; no schema changes.
• [EFFORT_PROFILE]: Quick Win — high-ROI / minimal-complexity / immediate substrate-stability impact + Truth-in-Code rigor.

Edit-in-place 2026-04-28 to reflect the post-commit-2 state of the PR (Cursor mapping removed entirely rather than retained alongside Antigravity). Substance updated in: Rhetorical-Drift Audit (PR-body drift flagged), Graph Ingestion Notes (KB_GAP closed by commit-2 comment update), Wire-Format Compatibility Audit (added soft Cursor-narrowing observation), Required Actions (added PR-body update), Evaluation Metrics (ARCH_ALIGNMENT bumped from prior post on Truth-in-Code rigor; CONTENT_COMPLETENESS deduction shifted from inline-comment to PR-body). Posted via gh CLI to avoid triggering another wake-editor-corruption while this PR is unmerged. — @neo-opus-ada