start_fine_tune

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

Beyond the annotations (readOnlyHint=false, destructiveHint=false, openWorldHint=false) the description discloses genuinely non-obvious behavior: training loops until max_steps optimizer steps over shuffled examples, every non-padding next token contributes to loss including chat prompt tokens with no assistant-only masking, bfloat16 runs under autocast with float32 trainable weights, and LoRA produces an adapter artifact. Those facts materially change how an agent should configure a run and interpret results, which is exactly what annotations cannot convey.

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 operative clause is front-loaded in the first sentence and every following sentence adds a distinct fact. The staccato fragments ('No assistant-only masking.') read abruptly but carry real content, so little is wasted beyond a small amount of terseness.

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?

An output schema exists, so return values need no explanation, and the description correctly signals asynchronous job-ID semantics. For a long-running, compute-heavy mutation, though, it omits lifecycle guidance — how to check progress or cancel, and what happens to an existing output directory/adapter on collision — which an agent needs before calling it.

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 0% across a nested 21-property TrainRequest, so the description carries the burden and only partly discharges it: it explains that max_steps counts optimizer steps (not epochs), what method=lora vs full does, how dtype=bfloat16 is implemented, and that LoRA output is an adapter. The remaining settings (learning_rate, batch_size, lora_rank, target_modules, gradient_accumulation, device, seed) get no added meaning, so most of the schema is left to name-based inference.

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 first sentence gives a specific verb and resource — 'Launch local LoRA or full-parameter training' — plus the immediate result ('return a job ID after validating inputs'). That wording is sufficient to separate it from siblings like start_evaluation, start_generation, and apply_weight_edits 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?

Usage is only implied: the launcher is clearly the entry point for a training run, but there is no explicit 'use this when X, use merge_adapter/job_status instead when Y' routing. It never tells the agent that a launched job should subsequently be polled with job_status or aborted with cancel_job, so the agent has to infer the workflow.

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