screenshot-diff

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

With no annotations, the description carries the full burden and it delivers extensively. It discloses the return contract ({ summary, diffPath, contextDiffPath }) and the coordinate normalization scheme ([0,1] screen locations matching describe coordinates). It also reveals important behavioral quirks: the fixed top status-bar band is ignored for both pixel and OCR comparisons, and live captures on physical iPhones are device-wide. It explicitly lists failure conditions (invalid sources, unreadable PNGs, unwritable outputDir, unreachable simulator-server/emulator backend), which is exactly the kind of behavioral context an agent needs and that annotations do not provide.

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?

The description is front-loaded with the core action, and each subsequent sentence adds distinct information: input modes, when to use, live-capture configuration, physical-device caveats, return format, ignored band, and failure conditions. It is dense but earned; every sentence carries a different fact. It loses a point only because it is longer than strictly necessary, and a couple of details appear twice in slightly different forms (the exactly-one live-capture rule appears in the body and again in the schema descriptions).

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 7-parameter comparison tool with no output schema and no annotations, this description is remarkably complete. It covers all input modes, the return value shape, coordinate conventions, the ignored status-bar band, device-specific behaviors, and failure modes. There is no obvious category of information an agent would need to invoke this tool correctly that is missing.

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?

The input schema already has 100% parameter coverage, so the baseline for this dimension is 3. The description adds significant semantics above that baseline: it explains the pipeline rule (exactly one of captureBaseline or captureCurrent; the common flow is baselinePath + captureCurrent), clarifies the role of outputDir (defaults to temp directory, artifacts still returned), and gives the physical-iPhone caveat that rotation is ignored. It also ties udid to backend resolution ('Always provide udid so the capture backend can be resolved'). These are real additions beyond the schema field comments, not repetition.

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?

The description opens with a specific verb-object pair ('Compare two PNG screenshots') and identifies the tool as a visual-diff operation distinct from the many sibling tools. It explicitly enumerates what the comparison covers (layout, spacing, color, typography, rendering, clipping, overflow, text rendering), so an agent can clearly distinguish this from screenshot or other capture tools. The sibling list contains many capture/gesture tools, but this one is unambiguously the diffing/comparison tool.

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?

The description explicitly states when to use this tool ('Use when stable before/after screenshots exist and the expected result is pixel-visible'), and provides concrete guidance on how to configure live captures ('set exactly one of captureBaseline or captureCurrent; use baselinePath + captureCurrent for the common visual-regression flow'). It also gives device-specific guidance (physical iPhones need no registered app, keep baselines per device model) and documents a key limitation (different aspect ratios fail). This is among the most usage-rich descriptions possible for a tool with no separate usage-hints annotation.

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