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

Annotations only give readOnlyHint=false, destructiveHint=false, idempotentHint=false, openWorldHint=true; the description goes well beyond them by disclosing that nothing is mutated yet, that a proposal id is returned, that the response shows per-record before/after state, that approval is human-gated via needsApproval, and that some actions would notify people in the customer's system. The "at most 25" batching limit is also stated. No contradiction with the annotations: readOnlyHint=false is consistent with a persisted proposal record.

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 when-to-use trigger and the not-written-yet caveat are front-loaded, which is right, but the workflow clause is a single run-on sentence with tangled phrasing ("say so first when something would be sent to the people in their system") that takes effort to parse. Several fragments lack a supplied verb. The content is substantive but the structure would benefit from being split into discrete sentences or steps.

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 proposal-generating mutation tool, it covers the essentials: confirmation gate, apply step, approval-link branch, batch size, and what the response contains. An output schema exists so return shape need not be spelled out beyond the before/after summary. It omits error/failure behavior (e.g., what happens with an invalid or already-unlinked pair), which keeps it from a 5.

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 both parameters (system, items and its nested fields) are already documented in the schema, including the get_ai_agent_options lookup and the 'never pick one yourself' instruction. The description adds only the restated batching cap ("Give one item per record, at most 25") which is already in the schema as minItems/maxItems. Baseline 3 applies because the schema carries the semantic load.

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 names a specific verb+resource ("The link between two records removed") and states the critical scope constraint that it is proposal-only ("Nothing is written yet... only the person can approve it"). It separates itself from read paths by naming them ("Not for reading (use search_records or get_record)"), though it never explicitly contrasts with the sibling propose_link_records. The clipped, near-telegraphic phrasing ("The link between two records removed") costs it a point on immediacy but the intent is unambiguous.

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?

It gives an explicit trigger ("Use when the person asks for this in a system whose AI agent may change records"), a full follow-up workflow (show the result to the person, warn when something would be sent to people in their system, ask for confirmation, then call apply_proposal, or share the approve link when needsApproval is true), and an exclusion routing reads to search_records/get_record. Alternatives are named with the condition that selects each.

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