CtrlK
BlogDocsLog inGet started
Tessl Logo

miloco-miot-scope

感知范围控制 — 管理 miloco 感知哪些家庭、哪些摄像头,每台摄像头的声音(是否参与感知),每台摄像头的「感知须知」(专属 prompt,补机位环境 / 关注 / 忽略以提升感知准确性),以及每台摄像头的「智能裁切增强」(Smart Crop,送模型的画面构图:裁主体 vs 送整幅)。用户说「只用/不用某家庭」「让 miloco 感知/别感知 某家庭或摄像头」「屏蔽某摄像头/把某摄像头从感知里去掉」「把某摄像头声音关了/打开」「XX 老误报(声音类)」「哪些家庭在用」时激活;也在用户吐槽**画面/视觉类固定误识**时激活——「门口摄像头老把电梯门当成我家门」「XX 老把 Y 认成 Z」「这台摄像头总把走廊的人当成家里人」「能不能告诉它忽略窗外的马路」等,用「感知须知」给该机位补指导;还在用户提**画面构图**类诉求时激活——「这台别裁了 / 只看整个画面 / 别老对着一块地方」「这台小目标看不清 / 远处的东西看不清」「智能裁切给这台关了 / 打开」等,用「智能裁切增强」(Smart Crop:裁主体 vs 送整幅)逐机位调。注意区分:开关摄像头设备本身(开机/关机/电源/录制)走 miloco-devices;感知引擎自身的开关/参数(分辨率档 / 全局裁切闸 / 感知窗口大小等)**没有 skill 覆盖**,miloco-perception 只有 perceive devices / logs / query / clear 四条命令(前三条是查询,clear 是清空感知流缓冲,都不是配置/开关),别往那儿转——正确入口(网页「设置」或 miloco-cli config set)与全局闸和逐机位闸的关系,见本 skill 正文。

73

Quality

91%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

SecuritybySnyk

Passed

No findings from the security scan

SKILL.md
Quality
Evals
Security

Quality

Content

82%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

An instruction-only skill with outstanding actionability and a genuinely well-sequenced diagnostic/triage workflow, including explicit validation checkpoints and feedback loops for batch operations. Its main weakness is conciseness — the routing facts and crop-troubleshooting detail are repeated across sections and could be consolidated.

Suggestions

Consolidate the repeated enumeration of miloco-perception's four commands (perceive devices/logs/query/clear) into one canonical statement; it currently recurs in the description, 工作方式, 何时激活, and the路由 sub-bullets.

Pull the 11-item adaptive_crop_fallback reason list out of the prose (lines 27-29) into a compact table or a one-level reference; the inline enumeration is the densest block and is explicitly disclaimed as incomplete, which both bloats tokens and risks staleness.

The裁切 troubleshooting tree (in_use/crop_in_use/全局闸, then connected, then reason=) reads as three nested paragraphs — restructuring it as a short decision-table would preserve the diagnostic power at lower token cost.

DimensionReasoningScore

Conciseness

Mostly efficient and free of generic concept-explanation padding, but several dense paragraphs (the裁切 troubleshooting tree at lines 25-29 and the repeated four-command enumeration of miloco-perception across sections) restate the same routing facts multiple times, which could be tightened without losing accuracy.

3 / 5

Actionability

Fully executable throughout: concrete CLI command forms, a complete command block, copy-paste-ready worked examples with realistic {code,message,data} envelopes, and exact field-by-field diagnostic recipes (in_use/is_online/connected/awake/crop_effective decision trees).

5 / 5

Workflow Clarity

The 误识→感知须知 flow is a numbered closed-loop with explicit checkpoints (locate did → draft 环境/关注/忽略 → 复述确认 → write → note 下窗生效), and the crop troubleshooting path has a structured diagnostic tree with explicit validation feedback (grep adaptive_crop_fallback, reason= values); validation against unknown dids is spelled out in the校验 table.

5 / 5

Progressive Disclosure

Well-organized with clear section headers (工作方式, 何时激活, 命令, 校验行为, 流程, 状态字段, 示例) and explicit pointers to out-of-skill surfaces (omni/prompt_builder docstring for the full 11-item fallback list, 网页「设置」, miloco-cli service logs); no bundle files exist, so all content is inline, which is appropriate for the skill size but leaves a couple of dense reference blocks (the fallback-reason enumeration) inlined rather than split out.

4 / 5

Total

17

/

20

Passed

Description

100%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A dense, highly specific description that maps four concrete capabilities to natural trigger phrases while explicitly routing edge cases away from sibling skills. Verbosity is its only real cost — it is far longer than typical descriptions — but the content quality is excellent.

Suggestions

Trim the routing/边界 details (miloco-devices vs miloco-perception vs no-skill-coverage) into the body; the description's job is what+when triggers, not the full conflict map, which bloats token cost on every skill lookup.

Move the global-gate / config-set mechanics out of the description; they are operational detail better placed in the body's裁切段落.

Consider tightening the trigger-phrase list to the highest-signal examples per axis to reduce redundancy (e.g. three near-identical composition triggers could collapse to one).

DimensionReasoningScore

Specificity

Enumerates concrete per-camera actions — manage which homes/cameras miloco perceives, per-camera mic participation, per-camera 感知须知 prompt (关注/忽略), and per-camera Smart Crop (裁主体 vs 送整幅) — with comprehensive coverage across the four feature axes.

5 / 5

Completeness

Explicitly answers both what (the four management capabilities) and when (concrete trigger phrases prefixed by 「…时激活」) with multiple concrete trigger clauses, satisfying both halves clearly.

5 / 5

Trigger Term Quality

Packs in natural user phrases users would actually say (「只用/不用某家庭」「屏蔽某摄像头」「XX 老误报」「门口摄像头老把电梯门当成我家门」「这台别裁了 / 小目标看不清」「哪些家庭在用」) including multiple synonyms and surface variants across visual-misID, audio, and composition complaints.

5 / 5

Distinctiveness Conflict Risk

Actively disambiguates from sibling skills (miloco-devices for device power, miloco-perception for query/clear, miloco-miot-admin for status) and even calls out the no-skill-coverage gap for engine params, giving it a clear niche with minimal conflict risk.

5 / 5

Total

20

/

20

Passed

Validation

93%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation15 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

metadata_field

'metadata' should map string keys to string values

Warning

Total

15

/

16

Passed

Repository
XiaoMi/xiaomi-miloco
Reviewed

Table of Contents

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.