Documentation
README
error — capture browser failure diagnostics
Capture artifacts when the browser shows a failure, then hand control back to the calling skill. Never stops the calling run by itself.
When to invoke
Any agent-browser snapshot that reveals:
- a "Something went wrong" heading
- "Authentication Error" or similar
- a blank page or unexpected redirect
- any content indicating the page failed to render
Steps
1. Resolve the slug
Use CARBON_WORKTREE from .env.local if present; otherwise the current
branch name:
SLUG=$(grep '^CARBON_WORKTREE' .env.local | cut -d= -f2)
[ -z "$SLUG" ] && SLUG=$(git branch --show-current | tr '/' '-')
2. Capture both artifacts
Save to .ai/scratch/e2e/ — it is gitignored (.ai/scratch/ in .gitignore).
Never save captures under docs/ (that's the documentation app, and it is
tracked).
mkdir -p .ai/scratch/e2e/${SLUG}
STAMP=$(date +%Y%m%d-%H%M%S)
agent-browser screenshot .ai/scratch/e2e/${SLUG}/${module}-${STAMP}.png
agent-browser snapshot -i > .ai/scratch/e2e/${SLUG}/${module}-${STAMP}.txt
${module} = a short name for the route/feature under test (e.g. accounting).
The .txt snapshot lets you debug element structure without reopening a browser.
3. Report and return
Log both file paths and continue — the caller decides whether to abort. Include the paths in the caller's final report so a human can open them.
Output
.ai/scratch/e2e/{slug}/{module}-{timestamp}.png— screenshot.ai/scratch/e2e/{slug}/{module}-{timestamp}.txt— element snapshot