Feature: Agent-team orchestration deep implementation

[chumyin]

Summary

Implement and validate an agent-team orchestration evaluation capability directly in this repository, including token-efficiency governance, topology benchmarking, and policy-gate verification.

Motivation

We need a practical, testable orchestration path for complex tasks that can be decomposed into parallel sub-work while controlling communication overhead and token spend.

Scope

• A2A-Lite message discipline and handoff constraints
• Topology evaluation (single, lead_subagent, star_team, mesh_team)
• Budget-aware degradation policies
• Deterministic JSON report + recommendation output
• Unit tests for gate pass/fail semantics

Validation

• Add and run tests locally
• Record outputs for gate pass/fail scenarios
• Sync evidence to Linear

Tracking

• Linear: RMN-290

[chumyin]

Implemented and validated in a clean worktree based on origin/main.

Added

• docs/project/agent-teams-orchestration-eval-2026-03-01.md
• docs/project/agent-teams-orchestration-eval-sample-2026-03-01.json
• scripts/ci/agent_team_orchestration_eval.py
• scripts/ci/tests/test_agent_team_orchestration_eval.py

Validation

• python3 -m unittest scripts.ci.tests.test_agent_team_orchestration_eval -v -> Ran 8 tests ... OK
• mesh baseline gate failure confirmed (--degradation-policy none)
• mesh recovery via auto degradation confirmed (--degradation-policy auto)
• all-budgets sample generated for low/medium/high

Status

Execution complete with reproducible evidence and policy-gate behavior.
Linear: RMN-290

[chumyin]

Follow-up implementation landed in repository code (not only scripts).

Core Rust module added

• src/agent/team_orchestration.rs
• exported from src/agent/mod.rs

Capabilities

• A2A-Lite message contract (A2ALiteMessage + HandoffPolicy)
• Team topology evaluation + budget/workload/protocol dimensions
• Auto/aggressive degradation under pressure
• Recommendation scoring and gate enforcement

Verification

• cargo test team_orchestration --lib -> 6 passed; 0 failed
• python3 -m unittest scripts.ci.tests.test_agent_team_orchestration_eval -v -> Ran 8 tests ... OK

[chumyin]

Deep implementation follow-up completed in repository core:

Added in Rust core (src/agent/team_orchestration.rs)

• conflict-aware DAG execution planner
• task budget allocator under run-budget pressure
• cycle detection and dependency validation
• ownership-lock-aware parallel batching

New verification

• cargo test team_orchestration --lib -> 9 passed; 0 failed
• python3 -m unittest scripts.ci.tests.test_agent_team_orchestration_eval -v -> Ran 8 tests ... OK

This extends the implementation from topology simulation into deterministic task decomposition and execution planning.

[chumyin]

Further hardening completed in core Rust planner:

Added

• validate_execution_plan for structural and budget consistency checks
• analyze_execution_plan for critical-path + parallel-efficiency diagnostics
• build_batch_handoff_messages for deterministic A2A-Lite batch handoffs

Verification

• cargo test team_orchestration --lib -> 12 passed; 0 failed
• python3 -m unittest scripts.ci.tests.test_agent_team_orchestration_eval -v -> Ran 8 tests ... OK

[chumyin]

Deep implementation update completed in clean worktree (issue-284) with fresh verification evidence.

Newly completed capabilities (repository core)

• Added end-to-end Rust API: orchestrate_task_graph(...)
  • topology evaluation -> recommendation -> planner config derivation -> conflict-aware plan -> validation -> diagnostics -> A2A-Lite batch handoff generation
• Added planner config derivation helper: derive_planner_config(...)
• Added communication token estimators:
  • estimate_handoff_tokens(...)
  • estimate_batch_handoff_tokens(...)
• Hardened plan validation with:
  • topological order checks
  • batch index consistency checks

Updated files

• src/agent/team_orchestration.rs
• src/agent/mod.rs
• scripts/ci/agent_team_orchestration_eval.py
• scripts/ci/tests/test_agent_team_orchestration_eval.py
• docs/project/agent-teams-orchestration-eval-2026-03-01.md
• docs/project/agent-teams-orchestration-eval-sample-2026-03-01.json

Verification evidence

cargo test team_orchestration --lib

Result: 17 passed; 0 failed

python3 -m unittest scripts.ci.tests.test_agent_team_orchestration_eval -v

Result: Ran 8 tests ... OK

This closes the research->implementation gap for token-aware agent teams with executable, test-covered orchestration primitives.

[chumyin]

Commit sync:

• Branch: issue-284
• Commit: 7dd71fcc
• Subject: feat(agent): add end-to-end team orchestration bundle

Fresh verification on this commit:

• cargo test team_orchestration --lib -> 17 passed; 0 failed
• python3 -m unittest scripts.ci.tests.test_agent_team_orchestration_eval -v -> Ran 8 tests ... OK