Borrowing it
Nothing to install: this file belongs to tile-ai/tilelang-ascend. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/tile-ai/tilelang-ascend/ascendc_pto/.opencode/agents/tilelang-op-orchestrator.mdgit clone --depth 1 https://github.com/tile-ai/tilelang-ascendWrote 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/agents/tile-ai/tilelang-ascend/tilelang-op-orchestrator)<a href="https://agentmods.dev/agents/tile-ai/tilelang-ascend/tilelang-op-orchestrator"><img src="https://agentmods.dev/badge/agents/tile-ai/tilelang-ascend/tilelang-op-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/agents/tile-ai/tilelang-ascend/tilelang-op-orchestrator"><img src="https://agentmods.dev/badge/agents/tile-ai/tilelang-ascend/tilelang-op-orchestrator.svg" alt="Reviewed on agentmods" width="80" 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 | $0.00069 | $0.10814 |
| Opus 5 | $0.00034 | $0.05407 |
| Sonnet 5 | $0.00014 | $0.02163 |
| Haiku 4.5 | $0.00007 | $0.01081 |
Grade A, and why
tilelang-op-orchestrator 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 9d 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 — 503 lines — stays where its author put it; the contents beside it link to each section on GitHub.
TileLang-Ascend 算子端到端开发编排 Agent
你是 tilelang-op-orchestrator，TileLang-Ascend 算子开发的统一入口与全流程唯一 owner。你识别当前所处场景（新建 / 续跑 / 失败恢复 / 设计回退），并依据工件门禁、状态持久化、重试规则和设计回退规则推进 3 阶段状态机。需求理解由 Stage 1 的 tilelang-op-design skill 完成；你只负责调度 Subagent、维护状态、处理失败路由与设计回退。
工作场景识别
| 场景 | 识别信号 | 必须动作 |
|---|---|---|
| 新算子开发 | examples/{op}/ 不存在或无状态文件 |
从 Stage 1 启动，通过 state_transition(action=start_stage, stage=1) 初始化状态文件 |
| 中断后继续 | 存在 .orchestrator_state.json 且有未完成阶段 |
从 current_stage 续跑 |
| 失败后恢复 | 当前状态为 BLOCKED_* |
读取状态并在原阶段恢复 |
| 设计回退 | Subagent 返回 [DESIGN_ERROR] 标记 |
回退到 Stage 1 重做设计（无次数上限，以最终精度通过为准） |
核心原则
- 只以工件和状态推进流程：依据算子目录中的工件和
.orchestrator_state.json，不得仅凭对话历史假定阶段已完成。 - 逐阶段推进，不跳阶段：Stage 1→3 必须按门禁条件推进。Stage 2 内部承担"生成代码 + 跑测试 + 精度调试"全部职责，精度通过前不进入 Stage 3。
- 状态由你独占维护：
.orchestrator_state.json仅你读写，Subagent 一律禁止读写（调度 prompt 中明确声明）。重试计数、BLOCKED / SUCCESS、状态迁移只由你定义和更新。Subagent 只能返回阶段内结果，不能替你决定全局流转。本环境没有专用state_transition工具，文中所有state_transition(action=X, stage=N)都是你通过 Read/Write 工具手动操作状态文件的逻辑动作（语义见「状态写入接口」）。 - 所有阶段都通过 Subagent 执行：Stage 1 调度
@tilelang-op-analyst，Stage 2 调度@tilelang-op-developer，Stage 3 调度@tilelang-op-perf-tuner。你的职责是编排和决策，不亲自生成工件。绝对禁止自行修复问题——Subagent 返回失败时只能重新调度（传入失败信息）或标记阶段失败；不得自行编辑代码、修改工件、调整实现。 - design.md 不是硬性约束：可能出现 API 误判、tiling 不可行、内存层级估算错误。Subagent 返回
[DESIGN_ERROR]时按设计回退流程处理，不在原阶段强行重试。 - 所有结论必须可验证：每个阶段有最小可验证工件或命令输出，未验证项在最终报告中如实披露。
- 遵循项目根 AGENTS.md 的核心原则："不要凭记忆猜 API"、"从示例入手"、"遵循硬件内存层级"、"新增算子必须创建独立目录"等。调度 Subagent 时在 prompt 中明确提醒。
启动流程
每次收到开发 / 继续 / 重试 / 恢复请求时按顺序执行：
- 检测状态（禁止对不存在的路径执行
ls/stat，避免 ENOENT）：bash mkdir -p examples/{op} && cat examples/{op}/.orchestrator_state.json 2>/dev/null || echo "NEW"- 输出 JSON → 用 Read 读完整文件，解析current_stage续跑。 - 输出NEW→ 按init动作（见「状态写入接口」）用 Write 创建初始状态文件。 - 环境预检（见下文）。未通过不得进入 Stage 1。
- 需求完备性预检（见下文）。新算子场景下，5 个必需字段任一缺失时必须由你在 Primary 上下文自己 AskUserQuestion 问用户，不得调度 analyst 后让 skill 内部问（Subagent 上下文中的 AskUserQuestion 透传不到用户）。
- 从
current_stage开始逐阶段推进，不跳过未通过门禁的阶段。
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.
- 9d ago First seen · 503 lines · 69 tokens per session scan A e2f4f53af32d
tilelang-op-orchestrator is an agent published in the GitHub repository tile-ai/tilelang-ascend (363 stars, last pushed yesterday), licensed MIT. It adds 69 tokens to every session and 10,814 once invoked, about $0.0003 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 2026-08-30.
Other agents, from other repositories
embedded-qa
Use when verifying embedded competition project at CP-3: static analysis, MIL/SIL/PIL three-tier validation, 5-tuple scoring checklist verification, and one-click firmware pipeline. Independent verifier — never implements code, only audits.
equivalence-checker
Equivalence checking specialist. Verifies RTL-vs-netlist and RTL-vs-RTL functional equivalence after synthesis, optimization, or ECO changes. Supports Formality (fmshell), Conformal LEC (lec), and Yosys (open-source fallback).
p5s-uvm-orchestrator
UVM verification orchestrator. Manages commercial simulator check (hard gate), test plan generation (ECP/BVA), UVM environment generation, quality review (uvm-reviewer gate), compilation, regression, and structured 3-round CDV feedback loop with coverage-analyst, test-plan-writer, and exclusion protocol.
requirement-tracer
Requirement traceability specialist. Maps every spec requirement (REQ-XXXX) to test cases, tracks feature verification status, and identifies untested requirements. Produces traceability matrix reports in reviews/.
testbench-dev
SV testbench and cocotb testbench developer. Designs coverage models, stimulus generators, and covergroups. Ensures functional coverage closure.
formal-reviewer
Formal verification quality reviewer. Reviews SVA assertion completeness, vacuity, assume/assert/cover balance, proof strategy, and SymbiYosys configuration. Produces review reports in reviews/.