一键在 Manus 中运行任何 Skill

agent-harness

星标2

分支0

更新时间2026年7月12日 07:33

Shared multi-agent harness state and coordination for Codex, Claude Code, and Antigravity. Use for repo resume/state, shared memory or active-task handoff, plan claims/recovery, fail-ledger checks, stale `.oms` cleanup, role/executor setup, local provider calls or write delegation, patch admission/landing, artifact provenance, and prior-session handoff.

安装

用 Codex 或 Claude 帮你安装复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。

在 Manus 中运行

来源

eightmm

eightmm/oh-my-setting

打开 GitHub 仓库查看创作者相关仓库

下载

在 Manus 中运行

Agent Harness

Keep cross-agent work reproducible while the current agent remains the owner.

Core Rules

On resume, run oms state --repo .; use oms init only for a fresh repo.
Keep secrets, machine/cluster details, raw logs, datasets, checkpoints, and private paths out of prompts and shared state.
The parent owns scope, plan approval, verification, patch landing, commit, push, release, and final synthesis.
Use an advisor only for irreversible/high-risk decisions, repeated failures, or a release go/no-go; routine completion does not require one.
Use oms gc for stale claims, workers, guards, runs, and retained artifacts. Do not repair .oms/ by hand.
Use oms agent-run --mode read for one read-only provider pass and --mode write for isolated worktree delegation. Do not rely on auto when the authority boundary is already known.
Land delegated changes through oms patch-land; a worker patch is evidence, not permission to mutate the main tree.
Provider subprocesses must remain harness children and must not recursively delegate or create state in their temporary worktree.
For a pre-authorized plan task, prefer oms plan-run over manually composing claim, delegation, review, and landing. It executes one task only, stops in review by default, and requires explicit --land for main-tree mutation.

Route to One Reference

Read only the reference needed for the request:

Memory, active task, prompt hooks, live edit guards: state-memory.md
Plan DAG, fail ledger, liveness, reclaim, GC: plans-recovery.md
Autonomous task intake, bounded progress, verification, and stopping: autonomy-loop.md
Strategy roles and task-scoped executor souls: roles-executors.md
Provider calls, delegation, artifacts, export/import, patch landing: delegation-artifacts.md
Capturing and loading a prior provider session: session-handoff.md

For experiment claims, run provenance, Slurm, or local GPU queues, use the research-method, slurm-hpc, or tsp-queue skill instead of loading harness details unrelated to the task.

Fast Paths

oms state --repo . --refresh-ci
oms agent-memory --repo . context
oms agent-task --repo . status
oms agent-plan --repo . ready
oms plan-run --repo . --to codex --next       # one task -> review
oms gc --repo .                 # dry-run
oms artifact-index --repo . unresolved

Report the provider called, useful conclusion, artifact or patch path, landing decision, and every failed or skipped verification.

同仓库更多 Skills

同仓库

chem-bio-ml

eightmm/oh-my-setting

Discovery-focused chem-bio ML guardrails for small molecules, molecular 3D and physics, proteins and variants, biomolecular interactions, peptides and antibodies, reactions and synthesis, generative design, cellular omics and phenotypic profiling, and biomedical networks. Use before writing, reviewing, evaluating, or training on RDKit/SMILES/SELFIES, QSAR/ADMET, quantum or force data, MMseqs2/MSA/protein language models, PDB/AlphaFold/DMS, docking/DTI/PPI, peptide-MHC or antibody data, reaction/retrosynthesis/yield data, molecular or protein generation, RNA/oligonucleotide/CRISPR, single-cell perturbation/Cell Painting, or biomedical knowledge graphs. Enforce domain splits, leakage checks, label semantics, baselines, metrics, calibration, applicability domain, and provenance.

2026-07-122

ml-training

eightmm/oh-my-setting

PyTorch training correctness for optimizers and schedules, distributed/DDP, variable-size masked losses, checkpoint save/load, CUDA runtime settings, and equivariant model checks. Use when writing or reviewing training loops, optimizer/scheduler setup, multi-GPU code, loss reduction, or checkpoints.

2026-07-122

peer-review

eightmm/oh-my-setting

Multi-agent code review. Use when the user explicitly requests independent verification, cross-agent/council review, or a release go/no-go or requested ML pre-training gate needs several independent reviewers. High-risk changes alone do not require this skill.

2026-07-122

research-method

eightmm/oh-my-setting

Hypothesis-driven loop for ML and scientific experiments. Use to turn a question into a falsifiable hypothesis, design the smallest disconfirming run, pre-register prediction/baseline/metric, interpret results, coordinate experiment claims, reproduce or compare runs, or trace an output to its run.

2026-07-122

slurm-hpc

eightmm/oh-my-setting

Slurm/HPC workflow helper. Use when working on clusters, partitions, nodes, sbatch/srun jobs, GPU/CPU allocations, queues, logs, checkpoints, or any resource-heavy command — e.g. "submit the job", "check the queue", "잡 돌려줘", "큐 확인", "GPU 몇 장", "학습 돌려줘" on a cluster; also digesting a long job log (job-digest, "로그 요약") and reconciling finished Slurm jobs into shared state (run-reconcile, sacct). Reads local cluster reference when generated.

2026-07-122

spec-interview

eightmm/oh-my-setting

Specification-first workflow for new projects, existing-repo onboarding, unresolved draft specs, and broad or architecture-shaping features. Detect project state before asking questions and resolve only choices that affect implementation. Clear bounded changes do not require an interview.

2026-07-122

name	agent-harness
description	Shared multi-agent harness state and coordination for Codex, Claude Code, and Antigravity. Use for repo resume/state, shared memory or active-task handoff, plan claims/recovery, fail-ledger checks, stale `.oms` cleanup, role/executor setup, local provider calls or write delegation, patch admission/landing, artifact provenance, and prior-session handoff.

Agent Harness

Keep cross-agent work reproducible while the current agent remains the owner.

Core Rules

On resume, run oms state --repo .; use oms init only for a fresh repo.
Keep secrets, machine/cluster details, raw logs, datasets, checkpoints, and private paths out of prompts and shared state.
The parent owns scope, plan approval, verification, patch landing, commit, push, release, and final synthesis.
Use an advisor only for irreversible/high-risk decisions, repeated failures, or a release go/no-go; routine completion does not require one.
Use oms gc for stale claims, workers, guards, runs, and retained artifacts. Do not repair .oms/ by hand.
Use oms agent-run --mode read for one read-only provider pass and --mode write for isolated worktree delegation. Do not rely on auto when the authority boundary is already known.
Land delegated changes through oms patch-land; a worker patch is evidence, not permission to mutate the main tree.
Provider subprocesses must remain harness children and must not recursively delegate or create state in their temporary worktree.
For a pre-authorized plan task, prefer oms plan-run over manually composing claim, delegation, review, and landing. It executes one task only, stops in review by default, and requires explicit --land for main-tree mutation.

Route to One Reference

Read only the reference needed for the request:

Memory, active task, prompt hooks, live edit guards: state-memory.md
Plan DAG, fail ledger, liveness, reclaim, GC: plans-recovery.md
Autonomous task intake, bounded progress, verification, and stopping: autonomy-loop.md
Strategy roles and task-scoped executor souls: roles-executors.md
Provider calls, delegation, artifacts, export/import, patch landing: delegation-artifacts.md
Capturing and loading a prior provider session: session-handoff.md

For experiment claims, run provenance, Slurm, or local GPU queues, use the research-method, slurm-hpc, or tsp-queue skill instead of loading harness details unrelated to the task.

Fast Paths

oms state --repo . --refresh-ci
oms agent-memory --repo . context
oms agent-task --repo . status
oms agent-plan --repo . ready
oms plan-run --repo . --to codex --next       # one task -> review
oms gc --repo .                 # dry-run
oms artifact-index --repo . unresolved

Report the provider called, useful conclusion, artifact or patch path, landing decision, and every failed or skipped verification.