aua_read_fields

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Adds substantial behavior beyond the readOnly/openWorld/idempotent annotations: OCR pixels are re-checked post-inference, missing or ambiguous values return NONE, decimal parsing is local, min_ocr_confidence is uncalibrated provider telemetry rather than accuracy, evidence_text is marked caller_provided_text and not verified as a phone read, and verified=false always requires Agent business verification. These are non-obvious trust and verification semantics an agent could not infer from the schema or annotations alone.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Is the description appropriately sized, front-loaded, and free of redundancy?

Highly information-dense with essentially no wasted claims, but it is delivered as a semicolon-chained wall of text with no grouping by parameter or mode, and jargon-heavy phrasing ('code-owned', 'tree', 'hashing'). The most decision-critical rule (auto vs tree vs screen_ocr) is front-loaded, but the rest requires re-reading to map to individual parameters.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a 13-parameter, no-output-schema tool with zero schema description coverage, the description supplies return semantics (NONE on ambiguity), verification obligations, environment prerequisites, and the OCR cost model. The remaining gaps are the undocumented tuning parameters (min_probability, min_margin, timeout_seconds) and the meaning of 'package', which an agent must guess or trial.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

With 0% schema description coverage, the description carries the parameter burden and does well for most: source modes, region limits in display pixels, ocr_masks excluding non-evidence from OCR and hashing, currency required for ambiguous symbols, sum_check validating same-currency Decimal sums, evidence_text provenance, and field kinds money/text/choice. However min_margin, min_probability, timeout_seconds, and package semantics are never explained anywhere, leaving four of thirteen parameters opaque.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Does the description clearly state what the tool does and how it differs from similar tools?

States a specific verb+resource ('Read-only Jev field judgments on current app text') and clarifies concretely with 'Extract code-owned money/text spans or classify supplied options, all fields in one request.' The term 'Jev' and 'code-owned' are unexplained jargon, but the second sentence makes the actual function intelligible and 'No visual clicks'/'no device actions' implicitly separates it from the aua_screen_action/tap family.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Does the description explain when to use this tool, when not to, or what alternatives exist?

Gives explicit mode-selection rules: source=auto uses the tree when text/expected_labels and money candidates suffice, otherwise captures a ROI and runs macOS Vision OCR; source=tree disables OCR, screen_ocr forces it, and OCR requires Xcode Command Line Tools. It also states a prerequisite ('Requires Jev enabled'). It never compares against sibling tools like aua_find or aua_observe, so there is no true alternatives guidance.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.