Skip to content

Instantly share code, notes, and snippets.

@gomesalexandre
Created May 3, 2026 12:42
Show Gist options
  • Select an option

  • Save gomesalexandre/b614c2e28b5f746f5c9019ae8b81eba9 to your computer and use it in GitHub Desktop.

Select an option

Save gomesalexandre/b614c2e28b5f746f5c9019ae8b81eba9 to your computer and use it in GitHub Desktop.
May 3 daily

May 3

My PRs (still open, authored today)

vultiagent-app

  • #380 SidePanel vault chip stops truncating mid-name - Text gets flex: 1 + minWidth: 0, parent row also minWidth: 0. "Dev - consider funds lost" no longer renders as "Dev - consider funds l..." with chevron crammed against the ellipsis. Bug 1 from yesterday's iPhone 16e dogfood, real before/after screenshot receipts. 5 source-shape regression tests
  • #381 skip redundant vault rehydration gate on AgentHome remount - useAgentOrchestration was initializing rehydratingVault to true and re-running loadFromStorage() on every mount even though AnimatedSplashScreen already hydrated the store. 5-7s of blank navy + spinner on every Scheduling-chip-tap-back-to-AgentHome. Two-line fix (lazy initializer + early-return), tap-to-AgentHome dropped from 5-7s to 1.5s. Bug 2 from the dogfood. 5 source-shape pins
  • #383 wire card_rendered into execute_* family - SendFlowCard, SwapFlowCard, ContractCallCard had ZERO card_rendered events firing, so the WS2b funnel was dark for roughly half of live tx volume. Mirrors the BuildTxCard contract (gate on awaiting_confirm, dedupe by messageId+surface, resolve chain dim from prep). Surfaced from a 3-day-old followup memo, verified the gap was still real on main before wiring. 2 regression-pin tests

Pushes on yesterday's PRs (still in flight)

agent-backend

vultiagent-app

  • #370 extractFee precedence - routed defaultDecimalsFor through resolveChainFromTxData so all fee-decimals lookups share the same network/chain/from_chain/blockchain precedence; formatFeeAmount helper at picoETH precision + sub-precision dust guard so fee row never silently reads "0 ticker" for non-zero spend
  • #371 useSendMessage onBeforeSend cancellation - swapped to ApprovalContext alias + async-aware onBeforeSend (signature widened, awaited, try/catch with isSendingRef released in outer finally so a throwing pre-send hook can't deadlock the send-lock)

vultisig-sdk

Reviews (no commits pushed)

agent-backend: #239 cosmos staking should-fix push - awaiting reviewer to push the bech32 checksum at schedule time + auto-pause story for withdraw_rewards empty-intersection

Issues filed (1)

  • vultiagent-app#382 fresh seed-import OTP has no skip / no back - hard-stuck without a real inbox. "Use a different email" technically rescues but is visually buried and destroys the in-flight vault. Tracking-issue not a quick PR: proper fix needs a vault.activationStatus schema migration + vault-list badges + settings re-entry + send-flow precondition checks. Body has full repro + 5-step fix sketch

Dogfood findings tally

  • Bug 1 (vault chip truncation): shipped as #380
  • Bug 2 (5-7s blank navy + spinner on Scheduling-back): shipped as #381
  • Bug 3 (prefilled prompt invisible during password gate): deferred, intent IS preserved in store but no UX feedback; needs a small intent-preview banner component
  • OTP screen trap: filed as #382
  • Credit Usage unauth: re-walked yesterday's #353 fix path, no regressions

Backend curl-replay battery

Found scripts/qa/curl-replay/ harness with 24+ canned conversations. Ran the safety-critical battery in direct-or mode against gemini-2.5-flash: 7 of 8 passed (prompt-injection, thrashing, fabricated-fee, self-send, wrong-chain, real-send-fabrication). The 1 failure is a known harness limitation: direct-or has no tool dispatch, so the LLM defaults to "I can't send transactions" instead of triggering the wire-side validator. Wire-side guards need --mode=backend with full ab + mcp-ts stack on DEV_SKIP_AUTH=true. Worth setting up next session for repeatable e2e receipts on the prompt-injection + breaker PRs.

Notes for tomorrow

  • Eight PRs sit CLEAN-and-MERGEABLE waiting on the next reviewer pass; Sunday was quiet
  • P1 followup (UserConfirmed / TxSigned / TxFailed / TxCancelled into useTransactionFlow.ts, ~120 LOC) is still open. #383 closed out the P2 (card_rendered on ExecuteCards)
  • OTP screen trap (#382) is real UX debt, worth queuing for a proper sprint with the schema migration + surrounding work
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment