fix(graphql): address all four tracked nits from PR #178's auto-review

[chrischall]

Summary

Follow-up to #178 (merged) — resolves every item left open in the auto-review-followup issue.

1. openEncryptedFrameDetailed's fromB64 calls sat outside every try — an iv like "A" passes BASE64_RE (which doesn't enforce length % 4) but still makes atob throw, out of a function documented as "never throws". Moved both calls inside the decrypt try, and reduced openEncryptedFrame to a thin rethrow-wrapper over openEncryptedFrameDetailed so the decrypt/parse/validate sequence exists in exactly one place.
2. content.ts's spoofed-graphql-res guard covered null/undefined but not arrays/primitives that assertObject also rejects — a spoofed data:[] or data:"x" from a same-window page script still reached assertObject(raw.data) and closed the bridge. Widened the guard to match assertObject's own object/array/null check, with parametrized tests over null, undefined, array, string, number, boolean.
3. capture-logger.ts's error string was OpenTable-specific ("open a restaurant page and retry") — the only vendor-specific string in extension-core's runtime output, relayed verbatim to every MCP using the graphql capability. Made generic.
4. graphql-query.test.ts only exercised ok:true — the actual bug the auto-review found (op:'graphql_query' failing validateInnerResponse) could only ever show up on the ok:false path, so this suite could not have caught it. Added an end-to-end case driving the real WS + encryption + validation + dispatch stack, asserting graphqlQuery() rejects, the extension WebSocket stays open, and a second call over the same connection still succeeds.

Closes #179

Test plan

• npm test — 1064/1064 passing
• npm run build — clean across all 7 workspaces
• npm run typecheck — clean
• No version bumps

🤖 Generated with Claude Code

[claude]

Claude finished @chrischall's task in 1m 28s —— View job

---

Reviewing PR #180

• Read CLAUDE.md conventions
• Read PR diff (git diff origin/main...HEAD)
• Review each changed file
• Post findings

Review

This is a clean, well-scoped follow-up that resolves all four items from #178's auto-review-followup issue (#179). Went through each change against the diff:

1. packages/protocol/src/seal.ts — openEncryptedFrameDetailed now moves both fromB64 calls inside the try, so a structurally-valid-but-undecodable base64 IV (e.g. "A", which passes BASE64_RE since it doesn't enforce length % 4) resolves to { stage: 'decrypt-failed' } instead of an uncaught throw. openEncryptedFrame is correctly reduced to a thin wrapper — verified the callers in background.ts/host.ts and all integration tests only do generic try/catch/.rejects.toThrow() on it, so the rethrow-wrapping (result.error instanceof Error ? result.error : new Error(...)) doesn't break any caller that depended on the old direct-throw shape.
2. packages/extension-core/src/content.ts — the graphql-res spoofing guard now matches assertObject's exact object/array/null check (packages/protocol/src/validate.ts:57-64: typeof x !== 'object' || x === null || Array.isArray(x)). Confirmed the mirrored check in content.ts is equivalent. New parametrized test (it.each over null/undefined/array/string/number/boolean) gives solid coverage of the previously-unguarded shapes.
3. packages/extension-core/src/capture-logger.ts — the OpenTable-specific "open a restaurant page and retry" string is gone; confirmed no other file in the tree still references it (grep -r "restaurant page" → no matches), so the test-file updates in validate.test.ts and content-graphql.test.ts are consistent with the source change.
4. packages/server/tests/integration/graphql-query.test.ts — new end-to-end case correctly drives the real WS/encryption/validation/dispatch stack, asserts the call rejects, the socket stays open, and a second call over the same connection succeeds with outboundSeq correctly continuing (not reset) to respect the strictly-increasing inbound-seq requirement.

No correctness, security, or convention issues found. Test/doc changes are consistent with the source changes throughout, and the PR's own claim of 1064/1064 passing lines up with what's in the diff.

Verdict: pass

[github-actions]

✅ Auto-review verdict: pass — PR correctly and cleanly resolves all four tracked nits from #178's auto-review, with solid test coverage and no correctness/security/convention issues found.