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 AgenticAIPlan/AgenticAISkills --skill paddleocr-ondevice-integration-navigatorgit clone --depth 1 https://github.com/AgenticAIPlan/AgenticAISkillsWrote 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/agenticaiplan/agenticaiskills/paddleocr-ondevice-integration-navigator)<a href="https://agentmods.dev/skills/agenticaiplan/agenticaiskills/paddleocr-ondevice-integration-navigator"><img src="https://agentmods.dev/badge/skills/agenticaiplan/agenticaiskills/paddleocr-ondevice-integration-navigator/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/agenticaiplan/agenticaiskills/paddleocr-ondevice-integration-navigator"><img src="https://agentmods.dev/badge/skills/agenticaiplan/agenticaiskills/paddleocr-ondevice-integration-navigator.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 | $0.00103 | $0.01891 |
| Opus 5 | $0.00051 | $0.00945 |
| Sonnet 5 | $0.00021 | $0.00378 |
| Haiku 4.5 | $0.00010 | $0.00189 |
Grade A, and why
paddleocr-ondevice-integration-navigator 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 11d 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 — 155 lines — stays where its author put it; the contents beside it link to each section on GitHub.
PaddleOCR 端侧集成导航
为 PaddleOCR 的 PP-OCR 系列模型端侧集成提供路线判断、集成路径收敛和 PoC 约束梳理。
适用场景
在以下场景使用本技能:
- 用户要把 PP-OCR 跑在 iOS / Android、嵌入式 Linux 或浏览器里。
- 用户要比较端侧集成路径,而不是只看单个示例仓库。
- 用户提到 ONNX、原生 SDK、厂商 NPU、离线推理或 PoC 指标,但需求仍属于端侧 OCR。
不适用场景
不要把本技能用于以下问题:
- 纯服务器侧部署,没有移动端、嵌入式或浏览器约束。
- 模型训练、蒸馏、微调。
- PP-StructureV3、PaddleOCR-VL、表格、公式、版面分析等非纯文字识别问题。
若用户的问题不在本技能范围内,明确指出并改查官方部署文档或相应产线文档。
收集输入
缺信息时只追问会影响路线判断的字段:
- 平台:iOS / Android、嵌入式 Linux、浏览器。
- 接入形态:原生应用、跨端宿主框架(React Native / Flutter 等)。
- 部署拓扑:纯端侧、端云混合,还是纯云。
- 输入形态:整页 / 拍照图、已裁切文本行、是否存在旋转或畸变。
- 语言与字符集:拉丁、CJK、多语。
- 预算:延迟、内存、包体、是否可用 GPU / NPU。
- 团队能力:原生 iOS / Android、RN / Flutter、Python、C++、仅 Web。
- 其他端侧工程约束:隐私、上传限制、弱网、缓存、重试、交互要求。
如果用户已经给出了这些信息,先复述约束,再继续判断。
工作循环
始终按下面的 loop 工作,不要在信息明显缺失时直接给结论:
- 先判断是否属于本技能范围。
- 如果不属于,立即说明原因,并引导到更合适的官方文档或其他技能。
- 检查关键信息是否完整。
- 以“部署拓扑、平台、接入形态、输入形态、语言与字符集、预算、团队能力”为最低判断集。
- 如果信息不完整,只追问当前最影响路线判断的问题。
- 一轮只问必要问题,不要一次抛出过多细枝末节。
- 优先追问会改变部署拓扑、平台路线、接入形态、运行时路线或模型变体判断的信息。
- 每收到一次用户补充,都先更新“已知约束”和“仍缺信息”。
- 仍缺关键信息时,继续追问。
- 已经足够判断时,停止追问,进入路由与输出阶段。
- 只有在最低判断集已经明确,或用户明确表示无法再提供更多信息时,才按照”路由请求“中的步骤查阅资料,给出结构化输出。
- 如果仍有缺口,需要在输出中把这些缺口标成
待验证,而不是假设补齐。
- 如果仍有缺口,需要在输出中把这些缺口标成
要求:
- 对于需要用户补充 / 明确的信息,调用 Agent 内置的提问工具提问,例如:
- 对于 Claude Code,使用
AskUserQuestion; - 对于 Codex,使用
ask_user_question; - 对于 Cursor,使用
Ask Question。
- 对于 Claude Code,使用
- 在最低判断集还不完整时,不要提前输出最终方案。
- 先追问,再判断,再输出;不要把追问和最终方案混在同一轮里。
路由请求
始终按下面顺序判断,不要跳步:
- 先读 references/task-shaping.md
- 判断是整页 OCR 还是仅识别已裁切行。
- 判断是否需要文档方向分类、文本行方向分类或文本图像矫正功能。
- 选择适用的模型变体。
- 再读 references/inference-topology.md
- 判断是纯端侧、端云混合,还是纯云。
- 明确哪些环节必须留在端上,哪些环节可以放到云上。
- 如果是纯云,确认是否存在需要单列说明的端侧工程约束。
- 如果是端云混合,先确定推荐拆分方式,再继续看平台和运行时。
- 判断目标是 iOS / Android、嵌入式 Linux 还是浏览器。
- 如果目标平台是 iOS / Android,再读 references/integration-hosts.md
- 判断是原生应用还是跨端宿主框架。
- 如果是跨端宿主框架,收敛
推荐接入形态,不能停在笼统描述。
- 再读 references/runtime-and-export.md
- 判断合适的集成路径,并给出建议。
- 需要做方案落地或对比验证时,再读 references/poc-and-validation.md
- 约束 PoC 指标、样本、对比口径和失败模式。
What ships with it
5 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.
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.
- 11d ago First seen · 155 lines · 103 tokens per session scan A 9ca09f2c49c5
paddleocr-ondevice-integration-navigator is a skill published in the GitHub repository AgenticAIPlan/AgenticAISkills (11 stars, last pushed 3mo ago), licensed MIT. It adds 103 tokens to every session and 1,891 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
orca-emulator-android
Android device and emulator control from inside Orca over adb, with the live device view in Orca's emulator pane. Use when driving an adb-connected emulator or phone on Windows, Linux, or macOS: booting AVDs, taps, swipes, typing, hardware buttons, rotation, app install and launch, runtime permissions, the…
android-tombstone-symbolication
Symbolicate the .NET runtime frames in an Android tombstone file. Extracts BuildIds and PC offsets from the native backtrace, downloads debug symbols from the Microsoft symbol server, and runs llvm-symbolizer to produce function names with source file and line numbers. USE FOR triaging a .NET MAUI or Mono Android app…
dogfood
Systematically explore and test a mobile app on iOS/Android with agent-device to find bugs, UX issues, and other problems. Use when asked to dogfood, QA, exploratory test, find issues, bug hunt, or test this app on mobile.
winapp-maui
Package and sign .NET MAUI Windows apps with winapp, resolving the resizetizer manifest dependency. Use when packaging or signing a .NET MAUI Windows app, building a MAUI MSIX or signed unpackaged build in CI, or fixing 'manifest contains unresolved placeholders ($placeholder$)' errors from winapp package.
react-native-ease-refactor
Scan for Animated/Reanimated code and migrate to EaseView.
apple-search-ads
When the user wants to set up, optimize, or scale Apple Search Ads (ASA) campaigns — including keyword bidding, match types, campaign structure, Creative Product Sets, CPP routing, and ROAS optimization. Use when the user mentions "Apple Search Ads", "ASA", "Search Ads", "Search tab ads", "Today tab ads", "CPT"…