The Verbose Flag That Kept Runs Alive
A one-flag compatibility fix that stops a CLI change from halting whole Gantry runs.
0 milestones 1 sprints
6m 28s total 1m 41s per sprint
1.3M tokens in 8.4k tokens out
$2.62 nominal cost
Gantry depends on Claude Code as a command-line harness, and its streaming path runs the CLI with `-p` and `--output-format stream-json` so the orchestrator can read events as they arrive. A newer Claude Code release made that flag combination invalid unless `--verbose` is present, which meant the process could fail before producing the stream Gantry expected. This job kept the compatibility fix small: add the required flag in the harness invocation and lock its placement with tests.
How this walkthrough is structured
Feature
What did the run build and ship?
The Claude harness still emits `--verbose` with its stream-json command, before user-supplied extra arguments, and the module still tests that requirement.
Build
How did Gantry structure the work?
The run was cut as a flat compatibility patch because the argument change, its placement rule, and its regression test all belonged inside the Claude harness boundary.
Claude Verbose Flag
0 agents 1m 41s wall time 3.6k tokens out $1.23 nominal cost
In the current tree, `src/engine/harness/claude.rs` still pushes `--verbose` immediately after `--output-format stream-json`, with a comment explaining the newer CLI requirement and the older CLI case. The exact argv test includes the flag, the extra-args test still checks that user arguments remain at the end, and a dedicated regression test asserts that stream-json invocations include `--verbose`. The archived run metadata files this job also touched are no longer present, but the harness behavior it introduced remains in source and tests.
The brief gave the agent a contained harness problem: adjust the argv Gantry builds for Claude Code without changing how user-provided escape-hatch arguments trail the command. That was small enough for a fresh worker because the behavior lived in one harness method and in the existing exact-argv tests beside it. The run facts show no repair, retry, or re-plan for this sprint, so the boundary did not come under recorded stress.