Problem
The current spec has no error handling in UI step sequences. VERIFY
tells an agent whether a flow succeeded or failed — but gives no
guidance on what to do next if it fails.
When step 4 fails, agents are left to guess. Common failure scenarios
with no current resolution path:
- A WAIT times out (element never appears)
- A VERIFY condition is false
- A NAVIGATE lands on an error page
- An INPUT is rejected by validation
Proposed addition
An ON-FAIL clause for steps or capabilities:
WAIT [data-agent-id="result-panel"] (max: 30s)
ON-FAIL VERIFY selector_exists [data-agent-id="error-message"]
Or a capability-level fallback declaration:
on-fail: retry-once | abort | notify-user
Source
Raised during implementation review. Without error paths, agent
behavior on failure is undefined and unpredictable across
implementations.
Problem
The current spec has no error handling in UI step sequences. VERIFY
tells an agent whether a flow succeeded or failed — but gives no
guidance on what to do next if it fails.
When step 4 fails, agents are left to guess. Common failure scenarios
with no current resolution path:
Proposed addition
An ON-FAIL clause for steps or capabilities:
Or a capability-level fallback declaration:
Source
Raised during implementation review. Without error paths, agent
behavior on failure is undefined and unpredictable across
implementations.