[rromenskyi]

When the judge confirms a goal, the agent now ends with a line like:

  ✓ goal met — worked 1h12m over 3 runs

What it counts:

• Worked is the working time of every run at the goal. It uses the existing runDuration, which already leaves out time spent waiting on an approval prompt. It is stored in goal.json (worked_ns, runs), so the total survives /goal go resumes and restarts.
• A goal replaced with different text starts its own count. A run with no standing goal counts toward nothing.
• In the TUI the line comes from a goal_done event emitted after the answer. Plain mode prints it after the answer.

Tests are in cmd/agent/goal_time_test.go:

• accumulation across a restart, plus resets;
• the duration format (42s, 4m05s, 1h12m);
• rendering the event;
• an end-to-end runTaskStreaming run with a fake model whose judge says DONE. It checks that goal_done carries this run's time.

A sabotage pass made 10 mutations, and every one failed a test. The plain-mode print in runOne has no test, because it writes straight to stdout.

🤖 Generated with Claude Code

[claude]

⚠️ Code review skipped — your organization has no extra usage available to pay for this review.

If your organization's extra usage balance is empty, an organization admin can add extra usage credits at claude.ai/admin-settings/usage. If its monthly spend limit was reached, an admin can raise it on the same page. If neither applies, contact Anthropic support.

Once extra usage is available, someone with write access to this repository can comment @claude review on this pull request to trigger a review.