A failed tool call is a count in the heading, not a red open run #95

Merged

mph merged 2 commits from cc-chat into main

2026-10-19 11:49:10 +00:00

claude commented

2026-10-19 08:41:31 +00:00

Member

Why

A tool call often fails on the arguments it was made with. The transcript marked that hard: the whole run opened, and cc-run[outcome="error"] cc-label coloured every row inside it — the calls that worked included. The machinery then stood in front of the conversation the reader came for.

What changed

A run with a failure folds like any other run. Only a picture still opens a run, because a picture a reader must click for is not shown.
A folded run says how many of its calls failed. The heading reads 21 tool calls · 1 call failed · 3m 1s · Bash 20, ToolSearch 1. The count is also attr(failed) on the element, beside outcome="error", so a project writes its own words.
The alarm's ink is on the count and on the answer that failed. Nothing else in the run is coloured.

Evidence

Measured on a real session, rendered by the server and read in a browser:

Check	Result
`cc-failed` colour	`rgb(185, 28, 28)` — the alarm token
The run's own label	`rgb(75, 85, 99)` — the page's quiet ink
Run folded	yes
Coloured rows with the run open	1 of 42 — the answer that failed

ui.test passed (exit 0) and __.checkFormat is clean.

Follow-up

The dashboard runs the released 0.7.0 and renders none of this. It needs a release, or a local snapshot, to show it.

🤖 Generated with Claude Code

https://claude.ai/code/session_01XTrxcxBQiNjXh79dN8Sh5x

## Why A tool call often fails on the arguments it was made with. The transcript marked that hard: the whole run opened, and `cc-run[outcome="error"] cc-label` coloured every row inside it — the calls that worked included. The machinery then stood in front of the conversation the reader came for. ## What changed - **A run with a failure folds like any other run.** Only a picture still opens a run, because a picture a reader must click for is not shown. - **A folded run says how many of its calls failed.** The heading reads `21 tool calls · 1 call failed · 3m 1s · Bash 20, ToolSearch 1`. The count is also `attr(failed)` on the element, beside `outcome="error"`, so a project writes its own words. - **The alarm's ink is on the count and on the answer that failed.** Nothing else in the run is coloured. ## Evidence Measured on a real session, rendered by the server and read in a browser: | Check | Result | |---|---| | `cc-failed` colour | `rgb(185, 28, 28)` — the alarm token | | The run's own label | `rgb(75, 85, 99)` — the page's quiet ink | | Run folded | yes | | Coloured rows with the run open | 1 of 42 — the answer that failed | `ui.test` passed (exit 0) and `__.checkFormat` is clean. ## Follow-up The dashboard runs the released 0.7.0 and renders none of this. It needs a release, or a local snapshot, to show it. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_01XTrxcxBQiNjXh79dN8Sh5x

claude added 2 commits

2026-10-19 08:41:31 +00:00

A call that failed is one coloured row, not a whole open run 25b08353c0

A tool call often fails on the arguments it was made with. A run opened
for every one of them put the machinery in front of the conversation,
and `cc-run[outcome="error"] cc-label` coloured every row in that run —
the calls that worked included.

The run now folds like any other, and the colour is on the answer that
failed, which the reader sees on opening the run. `cc-run[outcome=
"error"]` stays on the element for a project that wants folded runs
marked.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XTrxcxBQiNjXh79dN8Sh5x

A folded run says how many of its calls failed

CI / Compile (pull_request) Successful in 1m48s

Details

CI / Check Formatting (pull_request) Successful in 39s

Details

CI / Check Linting (pull_request) Successful in 1m24s

Details

CI / Run Tests (pull_request) Successful in 1m52s

Details

CI / Check Bundle Size (pull_request) Successful in 52s

Details

CI / Run Integration Tests (pull_request) Successful in 1m47s

Details

329e6615f5

A run that holds a failure now carries the count in its heading —
`1 call failed`, beside `21 tool calls` — and `attr(failed)` on the
element, like every other count a run reports. The alarm's ink is on
those words alone; the heading's own label and every row the run opens
onto keep the page's ink.

Measured on a real session in a browser: the heading reads
"21 tool calls | 1 call failed | 3m 1s | Bash 20, ToolSearch 1", the
count computes #b91c1c and the run's label #4b5563, the run is folded,
and of the 42 rows it opens onto exactly one is coloured.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XTrxcxBQiNjXh79dN8Sh5x

mph merged commit e25e75f3e5 into main

2026-10-19 11:49:10 +00:00

mph referenced this pull request from a commit

2026-10-19 11:49:11 +00:00

Merge pull request 'A failed tool call is a count in the heading, not a red open run' (#95) from cc-chat into main