When to Use
Use when the user wants to identify a plant from one or more photos, narrow down similar species, log a recurring houseplant or wild observation, or organize what to photograph next.
Architecture
Memory lives in ~/plant-identifier/. If ~/plant-identifier/ does not exist, run setup.md. See memory-template.md for structure.
CODEBLOCK0
Quick Reference
| Topic | File |
|---|
| Setup guide | INLINECODE4 |
| Memory template |
memory-template.md |
| Plant evidence checklist |
evidence-guide.md |
Scope
This skill ONLY:
- - identifies plants from visible traits in user-supplied images
- returns ranked candidates with explicit uncertainty
- asks for the next best photo when the signal is incomplete
- stores local observation notes only if the user approves
This skill NEVER:
- - declare a plant safe to eat, touch, burn, or medicate from chat alone
- guarantee species-level certainty when the plant lacks flowers, fruit, bark, or leaf details
- upload images or plant data to external services
Security & Privacy
Data stored locally if approved by the user:
- - activation and response preferences in INLINECODE7
- one note per saved observation in INLINECODE8
This skill does NOT:
- - make network requests
- give safety-critical edible or medicinal clearance
- write local files without user approval
Core Rules
1. Start with image quality and plant completeness
- - Check whether the image shows the whole plant, leaves, flower, fruit, stem, or bark.
- If the subject is distant, cropped, wilted, or mixed with other plants, ask for the most useful missing view first.
2. Return ranked candidates with evidence and uncertainty
- - Give one to three candidates with confidence bands: High 85-95, Medium 60-84, Low 35-59.
- For each candidate, say which visible traits support it and which missing traits keep it tentative.
- If the evidence only supports genus or family level, say so directly.
3. Use plant evidence in a fixed order
- - Open
evidence-guide.md before deciding. - Work from growth habit, leaf arrangement, margin, venation, flower structure, fruit or seed, stem or bark, then habitat context.
- Avoid jumping straight to flower color, which is often too generic by itself.
4. Ask for the next best plant photo, not more photos in general
- - Prefer whole-plant shot, leaf top and underside, node or stem view, flower front and side, fruit, and bark if relevant.
- Explain which missing trait would separate candidate A from candidate B.
5. Separate identification from safety claims
- - Plant identification can narrow likely species without proving edibility, toxicity, or medical use.
- If the user asks whether it is safe to eat, touch, or use medicinally, keep the answer conservative and provisional.
6. Keep memory around repeated observations, not noise
- - Save only durable preferences and approved observation notes.
- One saved entry should record date, location context, best match, confidence, and what evidence was missing.
- Do not write files unless the user approves local storage.
7. Say what could change the answer
- - Call out missing flowers, missing fruit, juvenile growth, pruning, indoor stress, and hybrid cultivars when they weaken certainty.
- Update the shortlist immediately if a better plant part is shown later.
Common Traps
- - Guessing species from leaf color alone -> many unrelated plants share the same color.
- Treating houseplant stress damage as a species marker -> environment gets mistaken for identity.
- Ignoring the leaf underside, stem nodes, or bark -> key differentiators stay hidden.
- Turning a tentative ID into edibility advice -> that creates avoidable safety risk.
Related Skills
Install with
clawhub install <slug> if user confirms:
- -
image - inspect and optimize plant photos before identification - INLINECODE12 - organize photo sets across repeated observations
- INLINECODE13 - broader plant care context once the plant is identified
- INLINECODE14 - improve close-up capture, lighting, and color reliability
Feedback
- - If useful: INLINECODE15
- Stay updated: INLINECODE16
何时使用
当用户希望通过一张或多张照片识别植物、缩小相似物种范围、记录室内植物或野外观察日志,或规划下一步拍摄内容时使用。
架构
记忆文件存储在 ~/plant-identifier/ 目录下。若 ~/plant-identifier/ 不存在,请运行 setup.md。目录结构参见 memory-template.md。
text
~/plant-identifier/
├── memory.md
├── observations/
│ └── YYYY-MM/
│ └── {entry-id}.md
└── exports/
快速参考
memory-template.md |
| 植物证据清单 | evidence-guide.md |
适用范围
本技能仅:
- - 根据用户提供的图像中可见特征识别植物
- 返回带有明确不确定性的排序候选结果
- 当信息不完整时,请求拍摄最佳补充照片
- 仅在用户同意后存储本地观察记录
本技能绝不:
- - 仅凭聊天内容宣称植物可安全食用、触碰、燃烧或药用
- 在植物缺少花、果、树皮或叶片细节时保证物种级确定性
- 将图像或植物数据上传至外部服务
安全与隐私
经用户同意后本地存储的数据:
- - 激活与响应偏好设置,存储于 ~/plant-identifier/memory.md
- 每次保存的观察记录对应一条笔记,存储于 ~/plant-identifier/observations/
本技能不会:
- - 发起网络请求
- 提供涉及安全性的食用或药用许可
- 未经用户同意写入本地文件
核心规则
1. 从图像质量和植物完整性入手
- - 检查图像是否显示整株植物、叶片、花、果实、茎干或树皮。
- 若拍摄对象距离过远、被裁剪、枯萎或与其他植物混杂,优先请求最缺失的有用视角。
2. 返回带有证据和不确定性的排序候选结果
- - 提供一至三个候选结果,并附置信区间:高 85-95、中 60-84、低 35-59。
- 对每个候选结果,说明哪些可见特征支持该判断,哪些缺失特征使其仍为暂定结论。
- 若证据仅支持属或科级别,请直接说明。
3. 按固定顺序使用植物证据
- - 在做出判断前,先打开 evidence-guide.md。
- 依次检查:生长习性、叶片排列、叶缘、叶脉、花结构、果实或种子、茎干或树皮,最后是生境背景。
- 避免直接跳到花色判断,因为花色本身往往过于笼统。
4. 请求拍摄最佳补充植物照片,而非笼统要求更多照片
- - 优先获取:整株照片、叶片正面与背面、节或茎干视图、花正面与侧面、果实及树皮(如相关)。
- 说明缺失的哪种特征能够区分候选结果A与候选结果B。
5. 将识别与安全性声明区分开
- - 植物识别可以缩小可能的物种范围,但无法证明其可食用性、毒性或药用价值。
- 若用户询问是否可安全食用、触碰或药用,回答应保持保守和临时性。
6. 仅保留与重复观察相关的记忆,而非冗余信息
- - 仅保存持久的偏好设置和经批准的观察记录。
- 每条保存的记录应包含:日期、地点背景、最佳匹配结果、置信度及缺失的证据。
- 未经用户同意本地存储,不得写入文件。
7. 说明哪些因素可能改变答案
- - 当缺少花、果实、幼龄生长、修剪、室内胁迫或杂交品种削弱确定性时,应明确指出。
- 若后续提供更佳的植物部位图像,立即更新候选列表。
常见陷阱
- - 仅凭叶片颜色猜测物种 → 许多不相关的植物可能具有相同颜色。
- 将室内植物胁迫损伤视为物种特征 → 环境因素被误认为身份特征。
- 忽略叶片背面、茎节或树皮 → 关键区分特征被隐藏。
- 将暂定识别结果转化为食用建议 → 造成可避免的安全风险。
相关技能
若用户确认,使用 clawhub install 安装:
- - image - 在识别前检查和优化植物照片
- photos - 跨多次观察整理照片集
- plants - 植物识别后提供更广泛的养护背景
- photography - 改善微距拍摄、光照和色彩可靠性
反馈
- - 如有帮助:clawhub star plant-identifier
- 保持更新:clawhub sync