Problem
During a step that ran for 241 seconds, there was zero output between [1/3] code-audit and the final failure message. For pipelines with tool-enabled steps that read dozens of files, this makes it impossible to know if the pipeline is progressing or stuck.
Current behavior
[1/3] code-audit # ← then silence for 4 minutes
✗ Step "code-audit" failed...
Suggested improvement
Stream intermediate progress from Claude's execution:
[1/3] code-audit
→ Reading app/src/lib/actions/auth.ts
→ Reading app/src/lib/actions/files.ts
→ Using Grep: "service_role"
→ Reading app/src/middleware.ts
...
✓ code-audit completed (241s, 45k tokens)
This could be done by parsing Claude CLI's stderr/stdout in real-time, or by using --verbose flag in the Claude CLI subprocess.
Impact
For pipelines expected to run 15-20 minutes, users have no way to distinguish "working" from "stuck" without this.
Problem
During a step that ran for 241 seconds, there was zero output between
[1/3] code-auditand the final failure message. For pipelines with tool-enabled steps that read dozens of files, this makes it impossible to know if the pipeline is progressing or stuck.Current behavior
Suggested improvement
Stream intermediate progress from Claude's execution:
This could be done by parsing Claude CLI's stderr/stdout in real-time, or by using
--verboseflag in the Claude CLI subprocess.Impact
For pipelines expected to run 15-20 minutes, users have no way to distinguish "working" from "stuck" without this.