hackathon-judge
/hackathon-judge
Ocenia projekt na hackathonie Meta według rubryki, prowadzi ostre pytania jury i przygotowuje trzyminutową prezentację oraz demo.
- Autor
- FrankieBiz ↗
- Repozytorium
- Faultline ↗
- Licencja
- Apache-2.0
- Ostatnia zmiana repozytorium
- 04.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 6
- Skrypty
- 1
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 04.11.2026, bez odwołań do wycofanych modeli i API.
Apache-2.0, otwarta: można używać i zmieniać, z zachowaniem licencji i informacji o zmianach.
- 1 skrypt w paczce. Przejrzyj je przed instalacją, bo Claude może je uruchomić.
Skrypty: scripts/judge.py
Co robi i kiedy
Ocenia projekt na hackathonie Meta według rubryki, prowadzi ostre pytania jury i przygotowuje trzyminutową prezentację oraz demo.
Wyzwalacze z opisu
Oryginalny opis (EN)
Judge, grill, and rehearse FaultLine for Meta's Global AI Developer Hackathon. Use when scoring the project against the Meta rubric (Impact 30%, Meta AI Tooling 25%, Technical Depth 20%, Demo 25%), running a hostile judge Q&A, rehearsing the 3-minute pitch, preparing the checkout-repair chaos demo, or generating a judging scorecard. Triggers on "judge this", "score the project", "grill me", "mock judging", "pitch rehearsal", "demo dry run", "scorecard".
Jak działa
- Reference Documents
- Ground Rules
- Phased Runbook
Podgląd SKILL.md
--- name: hackathon-judge description: Judge, grill, and rehearse FaultLine for Meta's Global AI Developer Hackathon. Use when scoring the project against the Meta rubric (Impact 30%, Meta AI Tooling 25%, Technical Depth 20%, Demo 25%), running a hostile judge Q&A, rehearsing the 3-minute pitch, preparing the checkout-repair chaos demo, or generating a judging scorecard. Triggers on "judge this", "score the project", "grill me", "mock judging", "pitch rehearsal", "demo dry run", "scorecard". --- # Hackathon Judge: FaultLine Act as a panel of skeptical Meta hackathon judges for FaultLine, the controlled browser QA and verified source repair platform. Score with concrete evidence, challenge architectural assumptions, and rehearse the 180-second pitch until it survives hostile technical interrogation. Every claim accepted or made must point to working code, an automated test, or an observable state in the running Inspector. ## Reference Documents | Document | Purpose | | --- | --- | | [references/meta_rubric.md](references/meta_rubric.md) | Official 100-point Meta Hackathon rubric: Impact (30%), Meta AI Tooling (25%), Technical Depth (20%), and Demo (25%) with scoring criteria and evidence anchors. | | [references/adversarial_grill.md](references/adversarial_grill.md) | 25 hostile questions across all four criteria, each backed by a grounded answer and repository evidence. | | [references/pitch_runbook.md](references/pitch_runbook.md) | The 180-second live pitch script with cue-by-cue timing, chaos trigger steps, and offline fallback maneuvers. | | [resources/scorecard_template.md](resources/scorecard_template.md) | Standard scorecard markdown template to record judging passes and track score evolution. | | [scripts/judge.py](scripts/judge.py) | Automated judging and verification CLI script to check test passes and evidence criteria. | ## Ground Rules 1. **Evidence or zero**: A criterion earns points only for verified implementation: working code, passing tests, or demonstrable UI state. Unimplemented plans score zero. 2. **Code beats documentation**: When documentation and code diverge, implementation is ground truth. Document any discrepancy as an explicit finding. 3. **Say what is deterministic**: Clearly separate deterministic execution (Playwright browser drivers, static diff checks, isolated sandbox execution) from model generation (Meta prompt repair candidates). Never mislead judges into believing deterministic rules were generated by LLMs. 4. **No live secrets on screen**: Never display raw API keys, tokens, or environment credentials during judging, pitches, or logs. ## Phased Runbook ### Phase 1: Verification Confirm all automated test suites pass before conducting judging: ```bash make check ``` A failing check caps the **Technical Depth** score at 10 points until resolved. ### Phase 2: Scoring 1. Copy `resources/scorecard_template.md` to a dated log file (e.g., `.agents/judging/YYYY-MM-DD-pass-1.md`). 2. Evaluate each criterion in `references/meta_rubric.md` against live repository evidence and running browser sessions. 3. Record itemized deductions, observed strengths, and final point totals. ### Phase 3: Adversarial Grill Conduct an interrogation using `references/adversarial_grill.md`. Presenter must answer each question cleanly in under 20 seconds using code-grounded proof. Mark each response as **held**, **shaky**, or **failed**. Remove unverified claims from the pitch immediately. ### Phase 4: Pitch Rehearsal Rehearse `references/pitch_runbook.md` strictly within the 180-second limit: - 00:00–00:30: The Problem (E-commerce silent checkout degradation). - 00:30–01:15: Chaos Trigger & Failure Reproduction (Run Chaos -> Playwright capture -> Inspector alert). - 01:15–02:15: Meta Model Source Repair (Single-file patch generation, isolated execution, regression suite verification). - 02:15–03:00: Impact & Production Guardrails (Deterministic safety boundary, business metric impact, pitch close). ### Phase 5: Remediation Rank deductions by impact and address top weaknesses before final submission.
Pliki w paczce (6)
- SKILL.md 4 KB
- references/adversarial_grill.md 3 KB
- references/meta_rubric.md 3 KB
- references/pitch_runbook.md 1 KB
- resources/scorecard_template.md 1 KB
- scripts/judge.py 2 KB
Wersje
W bazie nie ma innych skilli o nazwie hackathon-judge.
Migawka bazy z 13.11.2026 · wersja paczki 3d11e4517607 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100hackathon-strategist
Wspiera strategię na hackathonie od początku do końca: pomysły, wybór tematu, architekturę i MVP, pitch oraz demo, łącząc metodę lean startup z wzorcami zwycięstw. - 100hackathon-judging
Projektuje rubrykę i proces oceniania projektów hackathonu: kryteria i wagi, skalę punktów, pokrycie jurorów, normalizację ocen, rozstrzyganie remisów, konflikty interesów i ogłoszenie wyników. - 95hackathon-ai
Prowadzi przez hackathon z pomocą agentów AI: zakres, oś czasu, weryfikację, demo i zgłoszenie projektu w krótkim terminie. - 90Hedera Hackathon Submission Validator
Ocenia gotowość zgłoszenia na hackathon Hedera przed wysłaniem i wskazuje poprawki, które podniosą ocenę. - 90hackathon
Prowadzi przez hackathon: zrozumienie jury, minimalną specyfikację, testy najpierw, równoległe ścieżki, audyty i demo. - 100idea-judge
Ocenia pomysły na hackathon lub gotowe prototypy samodzielnie, w pięciu rundach, według prawdziwej rubryki wydarzenia i sprawdzania prac.