team-simulation
Orchestrate Project Aegis simulation and AI work: deterministic tick pipeline, sensors, engagement, logistics, policy evaluation, AI/delegation behavior, DOTS/ECS hot paths, replay verification, and m
SKILL.md.claude/skills/team-simulation/SKILL.md
GitHubTeam Simulation — Deterministic Sim / AI Orchestration
Use this skill for coordinated work on ProjectAegis.Sim, deterministic AI and
delegation behavior, DOTS/ECS simulation data, military parameter validation, and
replay reproducibility gates.
Phase 0: Resolve Mode and Review Level
Modes:
architecture— simulation assembly/API boundaries, system ownership, and ADR impact.tick-pipeline— fixed update order, command ingestion, policy, engagement, logistics, and logs.sensors— detection, classification, EW, contact state machines, and sorted pair iteration.engagement— fire-control, launch authorization, damage, DLZ, and abort reasons.logistics— magazines, fuel, readiness, supply, and data-driven constraints.policy— ROE/EMCON/WRA evaluation and delegation-to-sim policy migration.ai— autonomous controller behavior, decision purity, observed-state inputs, and seeded behavior.dots— DOTS/ECS, Burst, NativeContainers, BlobAssets, and hot-path layout.replay— order-log schema, replay hash, golden baselines, and divergence triage.accuracy— military parameter source review, authenticity/playability trade-offs, and balance.full-story— readiness → architecture → implementation handoff → tests → replay gate.- No mode — infer from the request; if ambiguous, ask one focused question.
Review mode:
- If
--review [full|lean|solo]is present, use it. - Else read
production/review-mode.txtif present. - Else default to
lean.
Team Composition
- military-simulation-architect — simulation design authority and orchestration lead.
- engine-programmer — core sim loop, pure C# systems, performance-critical code.
- ai-programmer — autonomous behavior, decision systems, perception/attention integration.
- determinism-engineer — seeded RNG, stable ordering, replay hash, wall-clock bans.
- unity-dots-specialist — DOTS/ECS, Jobs, Burst, NativeContainers, BlobAssets.
- sim-data-specialist — database-to-simulation snapshots and deterministic cache layout.
- simulation-parameter-analyst — source-backed military parameters and physical consistency.
- military-research-specialist — evidence/citation support for real-world systems.
- gameplay-mechanics-analyst — playability, balance, dominant strategy review.
- performance-analyst — frame/tick budgets and allocation review.
- qa-lead / qa-tester — test strategy, replay evidence, and regression coverage.
- technical-director — cross-layer architecture escalation.
Phase 1: Load Required Context
Read and summarize the minimum required context:
docs/architecture/architecture.md— layers, tick pipeline, ADR index, layer rules.- Relevant ADRs, especially simulation/delegation boundary, policy, order log, tick order, DOTS.
- Relevant GDD/requirements files for the feature or subsystem.
src/ProjectAegis.Sim/**,src/ProjectAegis.Delegation/**, and Unity adapter files in scope.- Existing tests under
src/ProjectAegis.Sim.Tests/**, delegation tests, and replay tests. - Data snapshot contracts if database-backed parameters are involved.
Report the deterministic boundary, owning agents, affected systems, and validation gates.
Phase 2: Select Workflow Pipeline
Architecture Pipeline
- Spawn
military-simulation-architectfor fidelity, ownership, and system boundaries. - Spawn
engine-programmerorc-sharp-architectfor pure C# API/assembly contracts. - Spawn
determinism-engineerfor ordering, RNG, and replay implications. - Consult
technical-directorfor dependency direction or ADR changes.
Tick / Replay Pipeline
- Verify fixed tick order and command ingestion order against ADR-004.
- Require stable keys for entity, side, contact, order, and event iteration.
- Run
/determinism-auditfor changed sim/controller paths. - Run or plan
/replay-verifywith golden order-log/world-hash evidence.
Sensors / Engagement / Logistics / Policy Pipeline
- Spawn
simulation-parameter-analystfor formulas and source-backed ranges. - Spawn
military-research-specialistwhen real-world evidence is needed. - Spawn
gameplay-mechanics-analystfor authenticity vs. playability review. - Spawn
engine-programmerorai-programmerfor implementation architecture. - Gate with deterministic tests and regression evidence.
AI / Delegation Pipeline
- Spawn
ai-programmerfor controller behavior and decision-system design. - Spawn
determinism-engineerto enforce pure functions of observed state, traits, and seed. - Coordinate with
lead-programmerfor API seams withProjectAegis.Delegation. - Require tests for edge cases, seeded behavior, and order-log impact.
DOTS Pipeline
- Spawn
unity-dots-specialistfor DOTS/ECS layout and Burst constraints. - Spawn
sim-data-specialistif data snapshots feed ECS runtime structures. - Run
/ecs-data-optimizationfor blittable layouts and NativeContainer safety. - Confirm no UnityEngine dependency leaks into pure sim assemblies.
Accuracy / Balance Pipeline
- Spawn
simulation-parameter-analystandmilitary-research-specialistin parallel. - Run
/simulation-accuracy-validationfor parameter authenticity. - Run
/gameplay-balance-analysisor/balance-checkfor playability concerns. - Document any real-world deviation and the approved reason.
Phase 3: Blocking Gates
Stop and ask before proceeding if any of these occur:
- Public sim API, tick order, order-log schema, or replay hash changes.
- New RNG streams or changes to seeded behavior.
- Unordered collection iteration on deterministic paths.
- Wall-clock reads or frame-time reads in simulation/controller logic.
- Per-tick database or UnityEngine dependency from
ProjectAegis.Sim. - DOTS/Burst design that is not blittable or allocates in hot paths.
- Military parameter changes without source/provenance or approved abstraction rationale.
Output
Produce a concise orchestration report:
- Mode and scope.
- Agents/skills invoked or recommended.
- Affected layers and source areas.
- Determinism, replay, DOTS, and accuracy risks.
- Required tests/evidence.
- Next approved action.