Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/ihsaan-ullah/auto-codabench/competition-designnpx skills add ihsaan-ullah/auto-codabench --skill competition-designgit clone --depth 1 https://github.com/ihsaan-ullah/auto-codabenchWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/ihsaan-ullah/auto-codabench/competition-design)<a href="https://agentmods.dev/skills/ihsaan-ullah/auto-codabench/competition-design"><img src="https://agentmods.dev/badge/skills/ihsaan-ullah/auto-codabench/competition-design.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00039 | $0.07747 |
| Opus 5 | $0.00019 | $0.03874 |
| Sonnet 5 | $0.00008 | $0.01549 |
| Haiku 4.5 | $0.00004 | $0.00775 |
Grade A, and why
competition-design scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 343 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Competition design
Source: Pavão et al., AI Competitions and Benchmarks: The Science Behind the Contests (2024). Chapter cites point to that book. This is the canonical, quotable source — when in doubt, cite it.
Use this as a decision tree. Resolve every section in order before launch — most failed competitions died at step 1 or 2, not the leaderboard.
0a. Live tensions in the competition-design literature
Surface these to the user when they hit. They are the dimensions on which thoughtful designers disagree, which makes them the most useful Phase A conversation accelerators.
| Tension | Sides | Where to read | Why it matters for the design |
|---|---|---|---|
| Fixed vs rolling test sets | Fixed test set is stable & comparable; rolling test set defeats memorisation by frontier models. | Pavão et al. (Ch. 3 §3.4); Roelofs et al. on benchmark drift | A NeurIPS reviewer will ask which you chose and why. |
| λ result-submission vs γ code-submission | λ is low-effort & community-friendly; γ is the only protocol that guarantees reproducibility and supports private test data. | Pavão et al. (Ch. 2 §2.4, Ch. 11 §11.2, Ch. 12 §12.1) | Decides whether you need a compute-worker, an ingestion program, and how anti-cheating works. |
| Multi-task aggregation: Borda vs mean-of-scores | Mean of normalised scores is interpretable but vulnerable to outliers; average-rank (Borda) is robust but discards magnitude. The book is opinionated for Borda. | Pavão et al. (Ch. 5 §5.6); Gibbard's theorem | Decides leaderboard ranking computation and what tie-breaking rule you write into competition.yaml. |
| Public-leaderboard size: small vs large fraction of test set | Small public leaderboard reduces overfitting risk but reduces feedback signal; large public leaderboard improves participant engagement but degrades into Goodharting. | Pavão et al. (Ch. 5 §5.1); Roelofs et al.; "Ladder" leaderboard (Blum & Hardt 2015) | Decides the public/private split ratio and the maximum daily submissions cap. |
| Adversarial-accuracy metric direction | For privacy/utility evaluation, ideal value is 0.5 (indistinguishable from real), NOT 1.0. Easy to set up wrong. | Pavão et al. (Ch. 4 §4.3) | If you mis-document this, participants optimize the wrong direction silently. |
| Detection-is-possible vs detection-is-futile (AI-text) | Sadasivan et al. (2023) argue detection accuracy → chance as generators improve; Krishna et al. (2023) + watermarking line argue this is escapable under specific assumptions. | oa:W4382349837 vs oa:W4385245221 |
If your competition is on AI-text detection, you need to take a position in the motivation section. |
| Single-objective vs multi-objective + weights | Multi-objective leaderboards (accuracy + latency + fairness) better reflect real deployment but require an explicit aggregation rule declared at launch — otherwise winners are unfalsifiable. | Pavão et al. (Ch. 4 §4.4, Ch. 12 §12.4) | If you have a multi-objective scoring goal, the aggregation rule IS the metric. |
| Open-generator vs closed-generator phases | Holding back generators for the final phase (open-generator) measures true generalization; freezing the generator set (closed) is fairer to participants who optimized for the announced set. | Pavão et al. (Ch. 5 §5.4); SemEval-2024 Task 8 design notes | This is the strongest distribution-shift lever and the dimension on which AI-text benchmarks diverge. |
| Code-sharing requirement: mandatory vs optional | CTF (Donoho 2017) says winner-shares-code is one of the four pillars. Practice varies: some competitions require it for the prize, some leave it optional. | Pavão et al. (Ch. 1 §1.1, Ch. 13 §13.2) | Decides what you can include in the post-comp paper / shared task series and whether winning solutions are reusable. |
| Pre-trained / foundation-model usage | Allowing foundation models speeds up entry but turns the competition into prompt engineering; disallowing them tests algorithmic novelty but excludes large fraction of the field. | Pavão et al. (Ch. 12 §12.3) | Particularly hot for NLP / vision competitions; usually best declared in the rules section. |
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 5d ago First seen · 343 lines · 39 tokens per session scan A af04c3168376
competition-design is a skill published in the GitHub repository ihsaan-ullah/auto-codabench (2 stars, last pushed 1mo ago), licensed MIT. It adds 39 tokens to every session and 7,747 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
agent-platform-rag-engine-management
Manage and query Agent Platform RAG Engine Corpora and retrieve grounded contexts using the Google GenAI SDK. Use when listing RAG corpora or files, inspecting a corpus, retrieving contexts, or generating content grounded in a RAG corpus. Do not use for standard database queries (use SQL/Spanner skills), Google…
agent-platform-model-registry
Agent Platform Model Registry Management. Use when you need to upload, list, describe, update, or delete machine learning models (and their versions) in the Agent Platform Model Registry. Don't use for model training, model deployment to endpoints, or managing non-Agent Platform models.
foundry-config-setup
Resolve missing setup caused by a hardcoded Foundry project endpoint or model in a sample. Use when a sample fails because it uses a placeholder/hardcoded projectendpoint (for example "https://your-project.services.ai.azure.com") or a hardcoded model instead of reading them from the environment.
google-cloud-solution-agentic-analytics-spark-knowledge-catalog
Discovers requirements and generates guidance to design and deploy a governed, secure agentic-analytics solution for data that's distributed across Google Cloud, other cloud providers, or on-premises. Data that's outside Google Cloud (such as data from Databricks, Snowflake, Salesforce, SAP, or Oracle systems) is…
training-check
Interactively monitor training metrics from the current Codex session, periodically checking WandB or fallback logs for NaN, divergence, plateaus, and broken runs.
nemo-automodel-launcher-config
Configure NeMo AutoModel job launches for interactive runs, Slurm clusters, and SkyPilot cloud execution.