tournament

TOURNAMENT — Competitive Coding with Penalty Scoring

N competitors. 1 judge. Winner's code ships.

Task: $0 Competitors: $1 (default: 5)

---

SCORING (Visible to ALL)

Criterion	Points	Penalty	Points
Correctness	40	Style nitpicks	-2
Elegance	25	Over-engineering	-3
Performance	20	Unnecessary complexity	-3
Completeness	15	Doesn't compile	-10
		False claims	-5

Tiebreaker: Correctness → Performance → First submitted

---

TEAM ARCHITECTURE

TOURNAMENT JUDGE (You — Orchestrator)
│
├─ Round 1: COMPETITION ($1 parallel competitors)
│  ├── competitor-1 ... competitor-N
│  └── GATE → all complete
│
├─ Round 2: JUDGING (1 agent)
│  └── tournament-judge
│  └── GATE → winner selected
│
└─ Round 3: IMPLEMENTATION (1 agent)
   └── winner-implementer
   └── Build + Test → SHIPPED | BLOCKED

---

<CRITICAL_EXECUTION_REQUIREMENT>

STEP -1 — Inherit Prior Findings: If <EXODIA_FINDINGS_CONTEXT> tag exists in session context, read .eight-gates/artifacts/findings.json. Inject relevant findings into competitor prompts as shared starting context — level the playing field.

YOU ARE THE JUDGE, NOT A COMPETITOR.

• DO NOT read files, write code, or fix issues yourself
• YOU ONLY: launch competitors, evaluate results, pick winner

1. Parse $1 for competitor count (default 5)
2. Launch $1 competitors in ONE message (Task tool, parallel)
3. Each works INDEPENDENTLY — no knowledge of others
4. Launch 1 judge to score all solutions
5. Launch 1 implementer for winning solution
6. Verify build + tests

YOUR NEXT MESSAGE: $1 Task tool calls. NOTHING ELSE.

</CRITICAL_EXECUTION_REQUIREMENT>

---

ROUND 1: COMPETITION

Launch $1 agents in ONE message. Identical prompt per competitor:

competitor-N (one per competitor)

subagent: feature-dev:code-architect | model: opus

TOURNAMENT — You are competing against others on the SAME task. Only the BEST wins.

TASK: $0

SCORING: Correctness(40) + Elegance(25) + Performance(20) + Completeness(15) = 100 PENALTIES: Style(-2), Over-engineering(-3), Complexity(-3), No-compile(-10), False claims(-5) TIEBREAKER: Correctness → Performance → First submitted

Write the BEST solution. Code must compile and pass tests. AVOID penalties. Explain WHY your solution should win.

Output: All code changes + approach explanation + why this wins