← 返回
未分类

SN94 Competition Poster

Use when creating, reviewing, or packaging SN94 BitSota competition proposals and replayable task repositories that fit the autoresearch backend format, incl...
用于创建、审阅或打包符合 autoresearch 后端格式的 SN94 BitSota 竞赛提案及可回放任务库,包括...
alveuslabs
未分类 clawhub v1.0.2 1 版本 100000 Key: 无需
★ 0
Stars
📥 183
下载
💾 0
安装
1
版本
#autoresearch#bitsota#latest#sn94

概述

SN94 Competition Poster

Use this skill to turn any measurable competition idea into a concrete SN94 task proposal and, when asked, a replayable public task repository skeleton.

Core Workflow

  1. Define the competition objective, input/output contract, hardware target, expected runtime, and what miners are allowed to edit.
  2. Convert that into the SN94 replay fields: repository_url, base_ref, allowed_patch_paths, setup_command, benchmark_command, result_path, and max_patch_bytes.
  3. Keep the validator generic. Put problem-specific scoring, datasets, and harness code in the public task repo. Hidden eval handles may come from the backend at replay time.
  4. Enforce narrow patch surfaces. Submissions must patch only allowed files, with no bytecode, cache files, generated artifacts, model blobs, wallet files, secrets, or repo-wide rewrites.
  5. Produce a poster-facing problem brief, an agent-facing implementation prompt, and an operator launch checklist.
  6. Tell the poster to submit the finished proposal as a GitHub issue in AlveusLabs/SN94-BitSota with title [Competition Proposal] , including the public task repo link.

When Creating A Task Repo

Read references/task-repo-spec.md for the task repo contract and problem quality rules. Do not assume a fixed problem type. Pick the simplest task-specific submission format that fits the problem and can be replayed safely.

Use scripts/create_competition_repo.py when the user asks for a starter task repo:

python scripts/create_competition_repo.py \
  --out /tmp/sn94-example-task \
  --slug example-frontier \
  --title "Example Frontier"

Review Checklist

  • The task repo can be cloned and checked out at a pinned immutable ref.
  • setup_command and benchmark_command run without network during reward-active validation unless the operator explicitly allows setup networking.
  • result_path is a JSON file with enough fields for the backend to score and audit the result.
  • allowed_patch_paths points only to miner-owned submission files.
  • max_patch_bytes is present and small enough to prevent model-smuggling and oversized patch attacks.
  • Public smoke tests exist, but release gating requires full local and testnet E2E replay.
  • Hidden validation can rotate by backend-provided env or manifest without changing the public repo.
  • The proposal states whether rewards are winner-take-all, pooled frontier, or another explicit policy.
  • The final answer includes the submission target: https://github.com/AlveusLabs/SN94-BitSota/issues/new?template=competition_proposal.md.

References

  • references/replay-contract.md: SN94 backend and validator replay contract.
  • references/task-repo-spec.md: task repo shape, examples, and scoring guidance.
  • references/marketplace-packaging.md: Packaging the same skill for AgentSkills, Hermes/HermesHub, and Claw/OpenClaw.

版本历史

共 1 个版本

  • v1.0.2 当前
    2026-05-28 13:45

安全检测

腾讯云安全 (Keen)

队列中

腾讯云安全 (Sanbu)

队列中

🔗 相关推荐

developer-tools

Github

steipete
使用 `gh` CLI 与 GitHub 交互,通过 `gh issue`、`gh pr`、`gh run` 和 `gh api` 管理议题、PR、CI 运行及高级查询。
★ 672 📥 324,995
ai-intelligence

Self-Improving + Proactive Agent

ivangdavila
自我反思+自我批评+自我学习+自组织记忆。智能体评估自身工作、发现错误并持续改进。
★ 1,373 📥 319,766
security-compliance

Skill Vetter

spclaudehome
AI智能体技能安全预审工具。安装ClawdHub、GitHub等来源技能前,检查风险信号、权限范围及可疑模式。
★ 1,223 📥 267,398