agentsclimarketplace

Ccdawn huawei nslb score loop

Skill CCDawn/codex-skills/skills/competition/ccdawn-huawei-nslb-score-loop

Chinese-first Codex skill pack for intent alignment, dynamic routing, lightweight development, code review, UI design, and AI research workflows.

Install
npx -y skills add CCDawn/codex-skills --skill ccdawn-huawei-nslb-score-loop

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 4 stars4 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use when optimizing, benchmarking, packaging, comparing baselines, handling online score feedback, or coordinating parallel agents for a Huawei Algorithm Challenge 37 NSLB workspace identified by its solver and score-loop artifacts.

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

5.0 KB, as published. Nobody here has run it

Huawei NSLB Score Loop

This is the Huawei NSLB project adapter for ccdawn-score-loop.

Use ccdawn-score-loop as the generic operating model, then apply the project-specific profile in references/huawei-nslb-profile.md when the request involves status, epoch work, workers, gate decisions, packaging, online feedback, failed-diff recovery, or project memory sync.

Profile

Project root: resolve from explicit `--project`, `CCDawn_HUAWEI_NSLB_ROOT`, or the current repository markers
Primary source: src/Solution.cpp
Tool: <codex-home>\skills\ccdawn-huawei-nslb-score-loop\scripts\score_loop_tools.py
Mutation space: references/mutation_space.json
Main ledger: docs/optimization-ledger.json
Online/offline weights: docs/online-offline-weights.json
Search graph: docs/search-graph.json
Attempt cards: docs/attempt-cards/
Failed diffs: docs/failed-diffs-index.json and docs/failed_diffs/

Known online-best anchor unless the project ledger says otherwise:

baseline: baseline-044-large-sparse-phase6-residual-target
online score: 344736301
source SHA256: 28b33d6d67d547fbc4415ec363a1c80db7a1f8e2fe1f308fd96a8a528a0441cb

BRT interface

  • Context Boundary: Huawei NSLB project root, current baseline hash, score-loop ledger/search graph, selected command/lane, and isolated worker workspace when applicable.
  • Output Contract: status, epoch, worker, gate, package, online feedback, or recovery artifact with machine-readable evidence.
  • Allowed Action: only profile-approved commands and lane workspaces; workers never mutate the main baseline, ledger, package map, or promotion state unless the parent gate explicitly does it.
  • Success Evidence: command output, child_result.json, attempt card, ledger/search-graph update, compiled diff, gate decision, or registered short submission package.
  • Stop Condition: source drift, stale baseline hash, overlapping ledger/pool write, missing child result, invalid package, ambiguous online feedback, or user pause.
  • Route Out: ccdawn-score-loop, project-specific command, launch/recover workers, parent promotion gate, online feedback wait, project memory sync, or BLOCKED with one required input.

统一调用契约

  • 只处理 BRT interface 范围;不匹配时回 ccdawn-brt 或更具体 owner,复合任务不吞其他 owner。
  • 用户可见内容默认中文,完成只报状态、产出、证据和剩余风险;代码、命令、路径、错误原文、API/协议、skill 名和枚举保留原样;Route Out 仅以 BRT interface 为准,末行写 下一步建议: <一个具体动作>

Required context commands

Before meaningful optimization, run or inspect the equivalent current outputs:

$project = if ($env:CCDawn_HUAWEI_NSLB_ROOT) {
  (Resolve-Path -LiteralPath $env:CCDawn_HUAWEI_NSLB_ROOT).Path
} else {
  (git rev-parse --show-toplevel 2>$null)
}
if (-not $project -or -not (Test-Path -LiteralPath (Join-Path $project "src\Solution.cpp"))) {
  throw "请通过 --project、CCDawn_HUAWEI_NSLB_ROOT 或当前仓库提供 Huawei NSLB 项目根目录。"
}
$codexHome = if ($env:CODEX_HOME) { $env:CODEX_HOME } else { Join-Path $HOME ".codex" }
$tool = Join-Path $codexHome "skills\ccdawn-huawei-nslb-score-loop\scripts\score_loop_tools.py"

python $tool status --project $project
python $tool doctor --project $project --min-score 0
python $tool drift --project $project

Do not rely on chat memory as the source of truth for score, source hash, active workers, or package mapping.

Command routing

Route these requests through the profile reference:

  • status, doctor, drift, restore online best;
  • prepare/run/dispatch/collect/close epoch;
  • lane pool update, worker recovery, child result validation;
  • failed-diff furnace, risk resurrection, search graph, attempt cards;
  • online feedback/update, submission registration, package gate;
  • dataset/proxy lane status and validation.

Use short upload names such as sub053.zip; store long metadata in docs/submission-map.json.

Worker rule

Workers must follow the generic ccdawn-score-loop contract plus these Huawei constraints:

  • main project is read-only for workers;
  • each lane edits only the assigned isolated workspace;
  • solver edits normally touch only src/Solution.cpp;
  • each edit lane must compile or return a concrete blocker;
  • each lane must run the smallest decisive first-kill before broad suites;
  • every worker must write child_result.json before completion.

For the detailed worker prompt, command router, mutation families, risk-resurrection rules, online calibration rules, and memory sync commands, load references/huawei-nslb-profile.md.

Completion

After meaningful Huawei score-loop work, update project memory with the lane competition-huawei-nslb unless the user explicitly says to skip it.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.