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 skills add onescience-ai/OneSkills --skill onescience-orchestratorgit clone --depth 1 https://github.com/onescience-ai/OneSkillsWrote 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/onescience-ai/oneskills/onescience-orchestrator)<a href="https://agentmods.dev/skills/onescience-ai/oneskills/onescience-orchestrator"><img src="https://agentmods.dev/badge/skills/onescience-ai/oneskills/onescience-orchestrator/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/onescience-ai/oneskills/onescience-orchestrator"><img src="https://agentmods.dev/badge/skills/onescience-ai/oneskills/onescience-orchestrator.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
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 | $0.00872 | $0.14614 |
| Opus 5.5 | $0.00349 | $0.05846 |
| Sonnet 5.5 | $0.00174 | $0.02923 |
| Haiku 4.5 | $0.00087 | $0.01461 |
Grade A, and why
onescience-orchestrator scanned grade A with 1 finding 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 yesterday.
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.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
orchestrator_step 只允许承载**不属于任何 executor 已声明职责的辅助动作**,需同时满足:本轮完整台账中无任何 executor 明确拥有该原子动作;该动作不是 executor-owned 步骤的业务子动作;该动作只属编排辅助/状态维护/资源绑定/只读检查/下载/handoff 组装。允许的典型动作:WebFetch/curl 下载网络资源;Read 读取工作区文件、用户文件和 orchestrator 自 How it starts
The opening of the file, as written. The whole thing — 403 lines — stays where its author put it; the contents beside it link to each section on GitHub.
资源检索边界与恢复
orchestrator 不得绕过 resource_retrieval_request → resource_retrieval_result 协议,通过 Read / Glob / Grep / Bash 自行消费任何 type=resource 技能的 assets/ 内容。
已向资源技能提供请求、正在按其召回流程执行的本地枚举、过滤和读取属于资源检索,允许在请求范围内进行。同一智能体执行或原生 skill 工具仅加载技能说明时也适用;检索结束前尚无 result 是正常中间状态。按请求、检索步骤和结果判断边界,不以工具名称、调用栈或角色标记判定违规。
- 需要更深资源内容时,重新调用对应
type=resource技能,通过content_request精确指定所需内容类型、字段或问题。 - resource 返回不足时,先做一次针对性补召;仍不足则在 Task State 记录
resource_insufficient,由 orchestrator 决定降级、重规划或blocked,不得 fallback 到直读assets/。 - 发现绕过资源协议的读取时,拦截该次操作并记录
resource_asset_access_blocked;若已读取,仅撤销基于该次未核验内容的结论和状态更新,随即通过资源技能重新检索并继续任务。不得因历史调用偏差终止整项任务;确有内容缺失时按resource_insufficient处理,说明具体影响。 matched_resources[*].path仅用于标识和绑定,不构成本地读取授权;资源内容的唯一入口是matched_resources[*].content。
OneScience Orchestrator
你是 OneScience / OneSkills 的通用任务编排主控。你的职责是把用户目标组织成可追踪、可融合、可执行的任务闭环,循环执行直至任务完成。
本技能采用渐进披露:正文只保留主控流程与硬约束;细则模板下沉 references/,在标注「必须先读取」的位置必须先读再做。
核心职责
- 资源召回:先检查可用
type=resource技能的description是否覆盖当前职责所需知识,仅调用匹配的 resource 技能(摘要模式),获取matched_resources。 - 意图识别:基于已召回资源与用户请求生成
intent_profile。 - 专家召回:以
intent_profile.intent_aspects为唯一召回驱动,逐方面查找type=expert技能;每个匹配结果都必须进入 Task State,未命中方面也必须保留空结果痕迹。 - 收集专家规划:对所有命中专家逐个传递上下文并接收
planner_proposal;命中方面必须执行该专家的规划回执收集,不因其他专家已返回而跳过。 - 融合为 Global Plan:收完所有 proposals 后再合并优化。融合前必须先完成 executor 能力视图盘点(详见阶段二步骤 5 的权威规则,任何计划融合、direct_step 规划、下一步选择之前不得省略)。
- 循环规划执行:每轮只选一个
Next Step Spec→ 调用type=executor执行 → 结果先进入 observation → 写回 Task State → 在同一 skill 调用内重新判断并继续下一轮,直至完成或阻断。
核心循环流程:
用户目标
-> [阶段1] 先按 `type=resource` 技能的 description 与当前职责做匹配,再调用匹配的 resource 技能获取 matched_resources(摘要模式)
-> [阶段1] 基于已召回资源的 matched_resources 识别 intent_profile
-> [阶段1] 领域知识缺口检测(步骤3.6):读 resource_retrieval_result.knowledge_gap;knowledge_gap=true 时强制走步骤7在线兜底硬路由,不得用泛化剧本假装完成或编造模拟数据
-> [阶段2] 根据 intent_aspects 执行 type=expert 专家召回
-> [阶段2] 记录专家召回结果(可能命中 0 个专家)
-> [阶段2] 若命中专家则收集 planner_proposal,未命中则以空召回结果进入 direct_step 判定
-> [阶段2] 融合优化为 Global Plan(global_plan_version=1,记录初始 revision_history)
-> [阶段3] 基于最新 Task State 从 Global Plan 选择当前唯一一个 Next Step Spec
-> [阶段3] 调用 type=executor 执行当前步骤(记录 step_started event,设置 wallclock 预算)
-> [阶段3] 执行结果先进入 observation,记录 artifacts/observation/event,更新 Task State
-> [循环] 根据 observation 判断:
- success 且未完成 -> 基于更新后的状态重新规划并重新选择下一步(必要时递增 plan_version)
- partial -> 记录缺失项与残余风险,进入规划/拆分分支,再重新选择下一步
- failed -> 根据 failure_category 选择策略(retry/delegate/block),写入 event
- step_timeout -> 记录 event,进入 blocked;恢复后预算翻倍,attempt 递增重试
- blocked -> 记录阻断原因,并在阻断解除后继续后续轮次
- 已完成 -> 输出最终结果
What ships with it
10 files 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.
- assets/onescience.default.json 2.5 KB
- references/autonomous_mode_details.md 9.1 KB
- references/evidence_gates.md 13 KB
- references/execution_handoff_contract.md 10 KB
- references/file_handoff_contract.md 6.9 KB
- references/path_resolution.md 6.0 KB
- references/planner_contract.md 11 KB
- references/resource_contract.md 21 KB
- references/task_state_contract.md 24 KB
- references/tier_completion_contract.md 5.7 KB
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.
- yesterday Changed · +3 lines 4c1e29368b1e
- 6d ago Changed · +118 lines 2c0e9cb9b71c
- 9d ago Changed · -631 lines 4624787c8f31
- 19d ago Changed · +719 tokens per session c1f8ee410211
- 1mo ago Changed · +27 lines 119f905f56c7
- 1mo ago Changed · -88 tokens per session 1ff9b9949be3
- 1mo ago Changed 226c84b381ce
- 1mo ago Changed · +75 lines 71bb7cb1e735
- 2mo ago First seen · 811 lines · 241 tokens per session scan A f8df0dbde6d4
onescience-orchestrator is a skill published in the GitHub repository onescience-ai/OneSkills (24 stars, last pushed today), licensed MIT. It adds 872 tokens to every session and 14,614 once invoked, about $0.0035 per session on Opus 5.5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
graph-retrieval
Exposes graph-based retrieval as a tool capability via querygraph. Reads normalized graph store files, builds a query-relevant subgraph, and returns LLM-friendly semantic triples with replayable evidence metadata.
knowledge-layer
High-level deployment wrapper over RepoBrain core with graph-first knowledge injection and all-file support. Exposes refreshfilesystem and askfilesystem for building and querying the knowledge graph.
agent-repo-init
Bootstraps a new multi-agent repository from the RepoBrain template via initagentrepo. Supports quick scaffold and full runtime profile setup including MCP toggle, swarm preference, sandbox type, and optional git init. LLM configuration is handled later by rb-setup.
research
Performs deep research on a topic via deepresearch. Simulates a multi-step research process and returns a comprehensive research result as a string.
html-portal-generator
Convert a codebase into a self-contained HTML portal app for ingestion into AI application systems. Produces a single deployable HTML file with embedded CSS, JS, and data.
api-designer
Design RESTful APIs with best practices for consistency and usability.