Skip to content

Commit fbad75c

Browse files
ianardeeianardee
authored
📝 add SKILL file (#439)
1 parent 0e8feb4 commit fbad75c

1 file changed

Lines changed: 174 additions & 0 deletions

File tree

‎SKILL.md‎

Lines changed: 174 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,174 @@
1+
# Mindee Python SDK
2+
3+
Use this skill for Mindee V2 integrations with the official Python SDK.
4+
5+
## Scope
6+
7+
- Use the official `mindee` Python SDK.
8+
- Focus on SDK-based integration patterns only.
9+
- Do not suggest direct HTTP calls, cURL, or non-SDK integrations.
10+
- Do not use undocumented SDK internals.
11+
12+
## Primary documentation
13+
14+
### SDK overview
15+
- https://docs.mindee.com/integrations/client-libraries-sdk.md
16+
17+
### Client setup
18+
- https://docs.mindee.com/integrations/client-libraries-sdk/configure-the-client.md
19+
20+
### Model parameters
21+
- https://docs.mindee.com/integrations/client-libraries-sdk/basic-model-configuration.md
22+
23+
### Load local files
24+
- https://docs.mindee.com/integrations/client-libraries-sdk/load-and-adjust-a-file.md
25+
26+
### Load remote URLs
27+
- https://docs.mindee.com/integrations/client-libraries-sdk/load-an-url.md
28+
29+
### Send files and URLs
30+
- https://docs.mindee.com/integrations/client-libraries-sdk/send-a-file-or-url.md
31+
32+
### Process responses
33+
- https://docs.mindee.com/integrations/client-libraries-sdk/process-the-response.md
34+
35+
### Handle errors
36+
- https://docs.mindee.com/integrations/problem-database.md
37+
38+
## Handling responses by model type
39+
40+
### Extraction
41+
- Use: https://docs.mindee.com/extraction-models/sdk-integration/extraction-result.md
42+
- Use this page for accessing dynamic fields from `response.inference.result.fields`.
43+
- Use this page for examples of `SimpleField`, `ObjectField`, `ListField`, confidence, and locations.
44+
45+
### Split
46+
- Use: https://docs.mindee.com/split-models/sdk-integration/split-result.md
47+
- Use this page for iterating over `response.inference.result.splits`.
48+
- Use this page for `document_type`, `page_range`, and optional chained extraction results.
49+
50+
### Crop
51+
- Use: https://docs.mindee.com/crop-models/sdk-integration/crop-result.md
52+
- Use this page for iterating over `response.inference.result.crops`.
53+
- Use this page for `object_type`, crop location, polygon data, and optional chained extraction results.
54+
55+
### Classification
56+
- Use: https://docs.mindee.com/classification-models/sdk-integration/classification-result.md
57+
- Use this page for accessing `response.inference.result.classification`.
58+
- Use this page for `document_type` and optional chained extraction results.
59+
60+
### OCR
61+
- Use: https://docs.mindee.com/raw-text-ocr-models/sdk-integration/ocr-result.md
62+
- Use this page for iterating over `response.inference.result.pages`.
63+
- Use this page for page text, words, and word polygon data.
64+
65+
## Default workflow
66+
67+
When answering questions, follow this order:
68+
69+
1. Initialize the SDK client.
70+
2. Configure `model_id` and other inference parameters.
71+
3. Load the input source.
72+
4. Optionally adjust the file before upload.
73+
5. Send with polling or webhooks.
74+
6. Process the response.
75+
7. Handle errors and retries.
76+
77+
## Answering rules
78+
79+
- Base answers on the documentation above.
80+
- Prefer documented SDK methods and patterns.
81+
- Use environment variables for API keys in production.
82+
- Reuse a client instance when possible.
83+
- Prefer polling for simple examples.
84+
- Prefer webhooks for production or high-volume workflows.
85+
- If a feature is not documented, say it is not officially supported.
86+
- If a user asks for code, keep examples minimal and working.
87+
88+
## Code sample rules
89+
90+
- Use Python examples only.
91+
- Use the official `mindee` package.
92+
- Show imports explicitly.
93+
- Include the exact documented class and method names.
94+
- Use placeholders like `MY_API_KEY`, `MY_MODEL_ID`, and `/path/to/file.pdf`.
95+
- Keep samples focused on one task.
96+
97+
## Preferred example topics
98+
99+
### Client initialization
100+
Use:
101+
- `Client(api_key="MY_API_KEY")` from `mindee.v2`
102+
- `Client()` with `MINDEE_V2_API_KEY` environment variable
103+
- Context manager: `with Client() as client:`
104+
105+
### Input loading
106+
Use:
107+
- `PathInput` — load from a file path
108+
- `BytesInput` — load from raw bytes (requires `filename`)
109+
- `Base64Input` — load from a base64 string (requires `filename`)
110+
- `FileInput` — load from a binary file object (`BinaryIO`)
111+
- `URLInputSource` — load from a remote HTTPS URL
112+
113+
### Sending documents
114+
Use:
115+
- `client.enqueue_and_get_result(ResponseClass, input_source, params)` for polling
116+
- `client.enqueue(input_source, params)` for webhooks
117+
118+
### Response handling
119+
Use:
120+
- `response.inference` — access the inference result
121+
- `LocalResponse(payload).deserialize_response(ResponseClass)` for webhook payloads
122+
- `response.is_valid_hmac_signature(secret_key, signature)` for HMAC validation
123+
124+
### File preparation
125+
Use:
126+
- `input_source.apply_page_options(page_options)` — trim or remove pages
127+
- `input_source.compress(quality=85)` — compress before upload
128+
- `input_source.fix_pdf()` — repair broken PDFs
129+
- `input_source.has_source_text()` — check for embedded text
130+
- `PageOptions(page_indexes=[...], operation="KEEP_ONLY", on_min_pages=0)`
131+
132+
## Avoid
133+
134+
- Direct REST examples
135+
- cURL examples
136+
- Manual authentication header construction
137+
- Bearer token examples for API keys
138+
- Non-Python examples
139+
- V1 examples unless the user explicitly asks for V1
140+
141+
## If the user is unclear
142+
143+
Ask for only what is needed:
144+
145+
- input type: local file or URL
146+
- delivery pattern: polling or webhook
147+
- model ID
148+
- runtime context: web server, worker, or script
149+
150+
## Output style
151+
152+
- Be concise.
153+
- Answer with runnable examples when code is requested.
154+
- Link to the most relevant doc section.
155+
- Do not overwhelm the user with every option.
156+
- Start with the documented default path.
157+
158+
---
159+
160+
# Agent Instructions: Querying The Documentation
161+
162+
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.
163+
164+
Perform an HTTP GET request on the documentation URL with the `ask` query parameter.
165+
Include `python+sdk+-+` at the beginning of the question to get answers specific to this library:
166+
167+
```
168+
GET https://docs.mindee.com/integrations.md?ask=python+sdk+-+<question>
169+
```
170+
171+
The question should be specific, self-contained, and written in natural language.
172+
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.
173+
174+
Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.

0 commit comments

Comments
 (0)