Tool calls, Arc style
Answer me with artifact · 2026-10-27 07:28

New version 0.5.10

You have 0.5.5. Found when this page was made.

What's new:

Run in a terminal:

npx skills update answer-me-with-html -y

Or tell your agent:

Update answer-me-with-html
See what's new

Tool calls, Arc style

A

Their clip (Elia, uiarc.dev author)

Arc tool-call group: Using N tools, per-row meta, inline retry, inline approval, Used 4 tools · 3.0s
B

Ours vs theirs

Moment Ours (ProcessBlock today) Arc
While running Plan line, last 2 steps, "+N earlier", live row, no frame Framed card, header "Using 3 tools" with stacked app icons, every row shown
Per row Icon, verb, detail, caret App icon, verb, detail, "4 results · 1.2s", green check, chevron
Tense Yes: Searching → Searched Same
Failure "Failed" pill; reason only after expanding Red row title, reason line under it, Retry button, header shows "1 failed"
Approval Separate ChatApprovalCard under the thread Row inside the group: Deny / Allow, "Sends as Arc agent to 48 people", header turns "Waiting for approval"
Done "Worked for 12s · 4 steps" + AppStack "Used 4 tools" + stacked icons + "3.0s"
After answer Regenerate in message actions "Run again" link under the answer

We already have the bones: app logos (#686), tense verbs, AppStack, receipts. What Arc adds is the frame, per-step timing, and failures + approvals living inside the group. ProcessStep has no per-step start/end time today, so timing needs a field.

C

No table: same features, no box

React · Two ways without the card. Click Retry, then Allow.
Which look, no card?
D

Five integrations

  1. Framed group with a live header. While running, the trail sits in a hairline card. Header: stacked app logos (reuse AppStack) + "Using 3 tools", then "Used 4 tools · 3.0s" when done. Every row shows, no "+N earlier" while it fits 6 rows.
  2. Per-step receipt: count + time + check. Add startedAt/endedAt to ProcessStep from the stream part timestamps. Right side of each row: "4 results · 1.2s" (count from receiptOf / appPreview) and a green check, spinner while live.
  3. Inline failure with Retry. The failed row reads "Could not …" in red. The error shows on a second line, no expand needed. A Retry button re-runs that one tool. The header says "1 failed".
  4. Approval as a row in the group. The pending ChatApprovalCard moves into the trail as the row it gates, with Deny / Allow. One line says what happens ("Sends as Raven to 48 people"), from approval-provenance. The header reads "Waiting for approval" and the border turns blue.
  5. "Run again" under the finished answer. A quiet link under the last answer calls the existing handleRegenerate. The phone and PWA rows get all five at phone sizing.
Which do I build (all surfaces + docs, one PR)?
E

Built: option B in the real component

The real component in dark mode: the rail, an approval row, and a failed row with Retry
The real component in dark mode: the rail, an approval row, and a failed row with Retry
What Where
Code allternit-ai #850 · feat/tool-calls-rail
Docs + map allternit-platform #1898
Checks tsrs clean; 130 test files pass; dark, light and phone width checked
Merge both PRs (deploys ai.allternit.com)?