Loading ToolBench
Digital Holographic Microscopy (DHM) reconstruction system with AI assistant integration. Provides tools for hologram loading, reconstruction, autofocus, quantitative phase imaging (QPI), depth mapping, and stage control.
28 tools with complete JSON schemas and descriptions present. Naming follows verb_noun convention (load_, set_, run_, get_, observe_, stage_, capture_, etc.). Descriptions average ~80 chars, meeting baseline minimums. However, parameter descriptions are sparse or missing entirely in many tools. Output schemas are undocumented, LLMs cannot predict return structure. Error handling is minimal; no recovery guidance. Security validation exists (path sanitization, clamp bounds) but not surfaced in tool descriptions. Composition is reasonable (single responsibility per tool), but many tools lack context about prerequisites and state dependencies.
Capture a single frame and optionally run reconstruction, autofocus, or QPI on it.
Capture a single frame from the camera (or current hologram in v1).
Compute a depth map by scanning z-range and applying a focus metric with local windowing.
Scan a z-range and return the top-k focus candidates ranked by a focus metric.
Retrieve the last N audit log entries (tool calls, parameter changes, results).
Retrieve the summary of the last result for a specific pipeline stage (recon, autofocus, qpi, or depth).
Query the current application state snapshot (loaded hologram path, reconstruction parameters, last results).
Output schemas completely undocumented. No tool describes what fields it returns, data types, or structure. LLMs cannot predict response format or chain tools effectively.
Irreversible operations (stage_move, stage_home, stage_jog, map_sample) lack confirmation or dry-run support. No recovery guidance if agent makes a mistake.
Error handling is minimal. Tool implementations return generic error dicts without actionable recovery guidance. LLMs cannot self-correct from failures.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2026-07-28+ | v2 |
Query the current LED intensity (0–100%).
Set LED illumination intensity (0–100%).
Load a hologram from the filesystem. Tilde expansion supported.
Scan a grid of (x, y) positions, capture frames, measure sharpness, and build a sample map.
Measure sharpness of the most recent frame using Laplacian variance.
Retrieve the amplitude field from the last reconstruction.
Retrieve the depth map from the last depth computation.
Retrieve the phase field from the last reconstruction.
Record a time-lapse sequence at the current (x, y) position, capturing frames at regular intervals.
Run autofocus to find the best z-position within a specified range using a focus metric.
Run quantitative phase imaging (QPI) to compute phase and refractive index maps.
Execute hologram reconstruction with current parameters.
Set reconstruction parameters (wavelength, pixel size, z-position, refractive index, mask radius, preprocessing options).
Switch the reconstruction reference mode (off, reference, or reference_free).
Close the shutter to block light.
Open the shutter to allow light through.
Perform a coarse-to-fine z-search to maximize sharpness at the current (x, y) position.
Query the current stage position (x, y, z in mm).
Home the motorized stage (move to origin).
Jog the stage by a relative offset (dx, dy, dz in mm).
Move the motorized stage to an absolute position (x, y, z in mm). Irreversible operation.
Parameter descriptions missing or generic. Many params (e.g., z_min_mm, z_max_mm, metric) lack context about units, valid ranges, or when to use each metric variant.
Tool 'capture_and_process' combines multiple responsibilities (capture + optional recon/autofocus/qpi). Should split into separate tools for composability.