Using the docs with an agent
These guides describe Clarity's user workflows. Agents can use them to plan tasks, explain results, or operate an available UI, SDK, or tool interface. The interface offered by the session determines which actions they can perform.
Find the right instructions
Start with the overview. For machine retrieval, use the Markdown index or page manifest. They link to plain Markdown copies of the same published guides and their canonical reader URLs. Use current task guides for instructions and the changelog for historical behavior and release status.
The manifest identifies the documentation target version and content hashes; it does not identify the application's deployed version. Confirm the target environment before using a Preview or Alpha workflow. Do not infer a tool name or API contract solely from the wording of a UI action.
Establish context before acting
Record the desired result and identify the project, workspace, image, and input layer. Resolve identifiers through the available interface rather than guessing them from names. An image, workspace source, processing layer, RGB result, dataset version, and model version are distinct entities.
| Task | Guide | Completion evidence |
|---|---|---|
| Import imagery | Upload data | Import finished; intended processing layer and metadata available |
| Create a color composite | RGB Explorer | Saved generation finished; input, channels, and processing recorded |
| Correct an image | Image versions | New image published; original retained; expected level and geometry |
| Prepare training data | Finalize dataset | Finalized artifact and documented split design |
| Train or apply a model | Training, Inference | Correct version and completed output; evaluation limits explained |
| Automate in Python | SDK | Installed version supports the method; returned resource verified |
Separate submission from completion
Import, correction, RGB generation, dataset finalization, training, and inference can continue after the dialog closes. Capture the returned job or resource identifier, inspect progress through the available interface, and verify the expected output before reporting completion.
A rendered RGB draft is unsaved. A queued correction is pending. A trained model is not automatically evaluated on independent test data. If work fails, inspect the message and existing state before submitting another copy.
Preserve interpretation
- Record wavelength units and whether a selector uses wavelengths or zero-based indices. Convert units explicitly: 1 µm equals 1000 nm.
- Distinguish geometry from processing level. Orthorectification changes the grid; assigning a sensor or level does not calibrate spectral values.
- Distinguish display stretch and overlay visibility from masks and processing settings that affect downstream data.
- Preserve originals and record the chosen corrected version as input.
- Use calibrated data and independent evaluation before claiming scientific accuracy from generated results.
For Clarity AI, see capabilities and tools. Discover tools offered by the session and follow its confirmation settings. Report result identifiers, outcome, relevant limitations, and links to saved outputs.