fix(core): 🐛 converge orphaned subagents and account final turns for judge-completed goals

[HayWolf]

Summary

• Add backstop to converge orphaned non-terminal run helpers (e.g. Judge subagents) to interrupted when a run finishes, preventing them from lingering at running forever after interrupt/cancel
• Account the final turn for goals that are judge-completed within the same run, so the finished turn count matches what was shown while the run was active
• Add helper_judge to the frontend helper kind formatter so Judge Agent shows correctly in the UI

Test Plan

• Verify orphaned helpers are converged on run cancel/interrupt
• Verify turn count stays consistent when Judge completes a goal
• Verify turn accounting is idempotent (no double count on re-evaluation)

🤖 Generated with TiyCode

[github-actions]

AI Code Review Summary

PR: #226 (fix(core): 🐛 converge orphaned subagents and account final turns for judge-completed goals)
Preferred language: English

Overall Assessment

No blocking issue was detected in the reviewed diff; keep focused regression testing before merge.

Major Findings by Severity

No major issues identified from the reviewed diff.

Actionable Suggestions

• Replace get_active() with a direct goal fetch by id to avoid stale turns_used in outcome payloads.
• Consider removing COALESCE for error_summary in interrupt_non_terminal_by_run to guarantee consistency between DB and events, or propagate the DB value into the SubagentFailed event.

Potential Risks

• If get_active() returns None, the frontend may show an incorrect turn count after a goal is completed.
• A helper with a pre-existing error_summary that is interrupted may display a different error message in the frontend than what is stored in the database.

Test Suggestions

• Integration test verifying SubagentFailed events for orphaned helpers.
• Goal lifecycle test for missing active goal after turn accounting.
• Run helper repo test for a helper that already has error_summary while being interrupted.

File-Level Coverage Notes

• src-tauri/src/core/agent_run_event_handler.rs: Adds a clean-up backstop that converges orphaned run helpers to interrupted and emits matching SubagentFailed events. The implementation is correct and properly idempotent.
• src-tauri/src/core/goal_manager.rs: Introduces idempotent turn accounting for the run that completes a goal, preventing a lost turn. The logic is sound, but the re-fetch via get_active() may lead to a stale payload in edge cases.
• src-tauri/src/persistence/repo/goal_repo.rs: New account_turn_if_unevaluated function is correctly written and idempotent; it uses a clear WHERE condition to avoid double counting.
• src-tauri/src/persistence/repo/run_helper_repo.rs: interrupt_non_terminal_by_run is a well-tested, transactional function that converges helpers safely. The COALESCE choice is the only small design consideration.
• src-tauri/tests/goal_lifecycle.rs: Expanded test correctly asserts that turn accounting occurs for the terminal run and is idempotent on re-evaluation.
• src/modules/workbench-shell/ui/runtime-thread-surface-helpers.test.ts: Adds unit test for the new helper_judge kind; straightforward and sufficient.
• src/modules/workbench-shell/ui/runtime-thread-surface-helpers.ts: Adds display mapping for helper_judge → 'Judge Agent', aligning with the backend.

Inline Downgraded Items (processed but not inline)

• None

Coverage Status

• Target files: 7
• Covered files: 7
• Uncovered files: 0
• No-patch/binary covered as file-level: 0
• Findings with unknown confidence (N/A): 0

Uncovered list:

• None

No-patch covered list:

• None

Runtime/Budget

• Rounds used: 1/4
• Planned batches: 1
• Executed batches: 1
• Sub-agent runs: 1
• Planner calls: 1
• Reviewer calls: 1
• Model calls: 2/64
• Structured-output summary-only degradation: NO

[github-actions]

Automated PR review completed.

• Findings kept: 0
• Findings with unknown confidence: 0
• Inline comments attempted: 1
• Target files: 7
• Covered files: 7
• Uncovered files: 0
  See the summary comment for detailed analysis and coverage details.

[github-actions]

Automated review completed for this PR diff. No concrete inline issue was selected after aggregation.