feat(subagent): ✨ Add parallel delegation

[jorben]

Summary

• Add agent_parallel for bounded concurrent delegation to explore, review, and read-only custom subagents.
• Add structured batch results with completed, failed, skipped, and batch status metadata.
• Add helper timeout, global helper concurrency limiting, prompt guidance, and frontend tool-name constant sync.

Test Plan

• cargo fmt --check --manifest-path src-tauri/Cargo.toml
• cargo test --locked --manifest-path src-tauri/Cargo.toml
• npm run typecheck
• npm run test:unit

🤖 Generated with TiyCode

[github-actions]

AI Code Review Summary

PR: #209 (feat(subagent): ✨ Add parallel delegation)
Preferred language: English

Overall Assessment

Detected 1 actionable findings, prioritize CRITICAL/HIGH before merge.

Major Findings by Severity

• HIGH (1)
  • src-tauri/src/core/tool_gateway.rs:640 - Tool execution semaphore can cause deadlock when terminal tools are also counted

Actionable Suggestions

• Unify tool and terminal semaphore acquisition order (always acquire tool semaphore first) to avoid deadlocks.
• Add abort-signal short-circuit inside run_parallel_delegate_task before consuming semaphore slots in the tool gateway.
• Refactor SubagentProfile to expose allowed_tools generically so parallel safety validation is future-proof.
• Rename copiedCopyTargetId to copiedTargetId for clarity.
• Add a dedicated semaphore or admission control for parallel helper subagent tool executions to avoid saturating the global tool_execution_semaphore.
• Run integration tests with 5 concurrent custom helpers, each using multiple tools, to validate that the system does not deadlock or timeout under maximum parallel load.

Potential Risks

• Deadlock under concurrent terminal tool use from parallel helpers.
• Parallel cancellation still burns limited execution slots.
• Future subagent profile types may bypass parallel safety checks.
• Resource contention between parallel subagent tool calls and main-agent tool calls could degrade performance or cause tool timeouts at high concurrency.

Test Suggestions

• Integration test: parallel batch cancelled mid-flight should release semaphore permits quickly.
• Integration test: multiple helpers using terminal-control tools under a max-concurrency smaller than 5.
• Unit test: parallel safety validation with a hypothetical non-Custom profile that has a tool list.
• Integration test: parallel subagent with 5 helpers, each calling read/search tools concurrently, verifying no tool timeout or semaphore starvation.
• Unit test: run_parallel_delegate_task behavior when tool gateway semaphore is fully acquired (simulate contention).
• E2E test: abort signal propagation during parallel execution to verify fail-fast and skip behavior.

File-Level Coverage Notes

• src-tauri/Cargo.toml: Minor dependency version update; no performance impact.
• src-tauri/src/core/agent_session_execution.rs: Parallel subagent execution is bounded by max concurrency (5) and task limit (5). No unbounded concurrency or memory issues. Sorting and serialization of small result sets is negligible.
• src-tauri/src/core/agent_session_tools.rs: Trivial change adding Parallel variant returning None; no performance impact.
• src-tauri/src/core/prompt/providers.rs: String changes only; no performance impact.
• src-tauri/src/core/subagent/mod.rs: No performance impact; module exports only.
• src-tauri/src/core/subagent/orchestrator.rs: Subagent tool executions now use a standard timeout instead of None, preventing runaway helpers. This may improve overall throughput by avoiding stuck tasks.
• src-tauri/src/core/subagent/parallel_contract.rs: Constants enforce bounded concurrency and task counts. No performance concerns.
• src-tauri/src/core/subagent/runtime_orchestration.rs: No performance impact; tool definition extensions only.
• src-tauri/src/core/tool_gateway.rs: Adding semaphores (max 5 concurrent tools, max 1 terminal control) reduces contention and improves stability. The slight reduction in parallelism is intentional and beneficial for resource fairness.
• src/modules/workbench-shell/ui/runtime-thread-surface-state.ts: New copy-state functions perform linear scans but operate on message arrays; negligible overhead for typical thread lengths.
• src/modules/workbench-shell/ui/runtime-thread-surface.test.tsx: No performance impact; unit tests only.
• src/modules/workbench-shell/ui/runtime-thread-surface.tsx: Assistant run copy state is derived via useMemo with a full scan of presentationEntries on each timeline change. For very large threads with frequent streaming updates this could cause noticeable recomputation, but it's unlikely in typical use. Consider incremental updates if profiling shows a bottleneck. (Potential but unconfirmed performance overhead for large threads with rapid updates.)
• src/shared/constants/tool-names.test.ts: No performance impact; test additions.
• src/shared/constants/tool-names.ts: No performance impact; constant addition.

Inline Downgraded Items (processed but not inline)

• None

Coverage Status

• Target files: 14
• Covered files: 14
• 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: 3
• Planner calls: 1
• Reviewer calls: 3
• Model calls: 4/64
• 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: 10
• Covered files: 10
• 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.

[github-actions]

Automated PR review completed.

• Findings kept: 0
• Findings with unknown confidence: 0
• Inline comments attempted: 1
• Target files: 10
• Covered files: 10
• 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.

[github-actions]

Automated PR review completed.

• Findings kept: 1
• Findings with unknown confidence: 0
• Inline comments attempted: 1
• Target files: 14
• Covered files: 14
• Uncovered files: 0
  See the summary comment for detailed analysis and coverage details.

[github-actions]

[HIGH] Tool execution semaphore can cause deadlock when terminal tools are also counted

Acquiring both a terminal-control semaphore and a general tool semaphore together can deadlock when multiple helpers compete for terminal tools under concurrency limits.

Suggestion: Use try_acquire with fallback, or acquire the tool semaphore first and only then acquire the terminal semaphore; or unify them into a single resource-permit model.

Risk: Under heavy parallel load with terminal tools, tool executions may hang until timeout instead of progressing.

Confidence: 0.85

[From SubAgent: general]