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

Goes well past the readOnly/openWorld annotations: it discloses the privacy model (free, private, offline, on-device), hard input bounds (~3,500 words, 2-50 labels), an output guarantee (the answer is always one of your labels, or several with `multi`), and an accuracy caveat with mitigation ('subtle or ambiguous text can be mislabelled, so spot-check'). That is unusually rich behavioral disclosure for a pure-inference tool.

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?

Four sentences, each doing distinct work: purpose first, then use cases, then limits, then exclusions. No filler or repetition, and the most decision-relevant information (what it does and where it applies) is front-loaded.

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?

Full picture for a five-parameter inference tool with no output schema: input constraints, output constraint, multi-label behavior, accuracy risk, privacy posture and scope exclusions are all present. Nothing an agent needs to call this correctly 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?

Schema description coverage is 100%, so text, path, multi, labels and instructions are all already documented. The description mostly restates those fields (3,500 words, 2-50 labels, text-or-path) and only marginally adds the framing that the answer is drawn from the label set. Baseline 3 is correct when the schema carries the parameter burden.

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 and resource: pick the best label for a text from a caller-supplied list, running on the on-device Apple Foundation Models. The scope exclusion ('Not for code, maths or reasoning') sharpens the boundary, but it never names the sibling tools (extract, ocr, summarise), so routing among them relies on inference.

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?

Concrete use cases are given ('triage: sorting tickets, emails, comments or log lines into categories') plus an explicit when-not ('Not for code, maths or reasoning'). It lacks any named alternative tool or a rule for choosing between this and summarise, so it falls short of full when/when-not/alternatives coverage.

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

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

Annotations already declare readOnly/non-destructive/non-open-world, so the bar is lower, yet the description adds substantial context: free/private/offline on-device execution, ~3,500-word input cap, the risk that absent fields take wrong values from nearby text, and the unreliability of nested schemas. This is exactly the behavioral context annotations cannot express.

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?

Dense but front-loaded: capability first, then good-use cases, limits, accuracy caveats, and schema constraints. Every sentence adds operational value, though it packs enough caveats that it reads as a block rather than distinct scannable points.

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?

Covers input sourcing, size limits, accuracy expectations, and schema restrictions for a tool with no output schema, and annotations cover the safety profile. An agent has everything needed to call this correctly and to know when results should be verified.

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?

Schema coverage is 100%, so the baseline is 3. The description goes beyond the schema by enumerating the JSON Schema keywords the tool actually honors (type, properties, required, items, enum, const, description, minItems, maxItems) and clarifying that text and path are alternates, which the schema only implies via defaults.

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 (extract) plus resource (fields into JSON matching a JSON Schema) and the engine (on-device model for Apple Foundation Models). It is clearly distinguishable from classify, ocr, and summarise, which all transform rather than populate a caller-supplied schema.

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 concrete when-to-use context (names, dates, amounts, IDs in short documents, emails, logs), an explicit exclusion ('Not for code or reasoning'), and input-size guidance. It never names a sibling tool as the alternative, so routing to classify/ocr/summarise is left to inference.

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

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

Annotations already cover the safe/read-only and closed-world profile, and the description goes well beyond them: on-device/free/private/offline processing model, an accuracy caveat ('small or unusual text can be misread; check numbers that matter'), and the exact return shape (line-by-line text or a prompt answer). That is the kind of behavior an agent needs before deciding to trust the output.

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?

Three dense sentences, front-loaded with the action and inputs, then return behavior, then caveats. No filler; each clause carries a constraint or capability an agent must know.

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?

With no output schema, the description still describes the return value (line-by-line text or a prompt-driven answer) and the failure mode (misread text). For a two-parameter read tool with full schema coverage and annotations, nothing material 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?

Schema coverage is 100%, so baseline is 3, but the description adds real meaning the schema lacks: accepted formats are constrained (PDFs explicitly unsupported, absolute path required) and the empty-prompt case is spelled out as 'return all the text' rather than an answer.

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 and resource ('Read the text in an image file') with example inputs and the underlying engine. It also distinguishes itself from siblings by excluding code review and reasoning, which would otherwise collide with classify/extract/summarise.

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?

Explicitly states when the tool applies (image files, absolute path) and key exclusions ('PDFs are not supported', 'Not for code review or reasoning'). It does not name a specific sibling to use instead for those excluded cases, so the routing guidance is clear but not fully closed.

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

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

Annotations cover only the safety profile (readOnly/openWorld/destructive), while the description discloses free/private/offline operation, concrete input ceilings (~30k words vs ~1000 dense log lines), the long-input split-and-combine behaviour with its latency and detail-loss cost, and the accuracy caveat ('can miss or distort details; check anything important'). This is unusually rich behavioral context beyond structured fields.

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?

Purpose is front-loaded in the first clause and every following sentence carries distinct information (cost, scope, limits, path hint, degradation, exclusions, caveat). It is dense and semicolon-heavy rather than cleanly chunked, which is the only thing keeping it below a 5.

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?

With no output schema, the description carries the return-shape burden and does so ('a paragraph, or at most 7 bullets'), matching the `length` enum. Input limits, degradation behaviour, the two input modes and the accuracy caveat are all present, so an agent has everything needed to call it correctly.

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?

Schema coverage is 100%, so the baseline is 3, but the description adds genuine decision value: 'Pass `path` to a text file instead of `text` to avoid reading it yourself' explains the cost tradeoff between the two mutually exclusive inputs. It also reinforces the enum contract ('at most 7 bullets', 'a paragraph'), though it does not explain the `focus` parameter.

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 ('summarise') plus the resource and the execution substrate ('on-device model for Apple Foundation Models'), and immediately frames the output as a gist rather than a full record. An agent can distinguish this from classify/extract/ocr without opening any schema.

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?

Explicit when-to-use ('condensing logs, documents, notes and transcripts into a gist') and explicit when-not ('Not for code, maths, reasoning or facts', 'not a full record'), plus a precondition ('filter bigger logs first (grep the errors)'). Nothing is left to inference.

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