perf_history

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

Annotations only declare readOnlyHint and openWorldHint, so the description carries the real behavioral load and does so: it states the sort order ('oldest first'), the cross-platform comparability caveat ('Runs from Windows and Metal in one series are flagged: their GPU ms and memory do not compare'), and the noise-floor interpretation rule. These are non-obvious traits an agent could not infer from annotations or schema.

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?

Verb and resource are front-loaded and nearly every clause carries information (ordering, metrics, delta, noise threshold, platform caveat, skip semantics). It is a single dense paragraph whose parentheticals ('the start-up', the translation-layer example) border on redundant given the schema, but nothing is padding.

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 an output schema present, return values need not be explained, and the description still covers ordering, comparability caveats, and the interpretation rule that an agent would otherwise miss. The remaining gap is the limit/sort interaction (capped at the newest N but reported oldest-first), which affects how a caller reasons about truncation.

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%, and the description largely mirrors it — skip_s as 'the start-up' seconds with recomputed metrics is already spelled out in the schema, and prefix semantics are already there. The one genuinely additive phrase is ordering ('oldest first'), which is not a parameter but interacts with limit ('newest runs to keep, default 50'); the description does not explain that tension.

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?

Opens with a specific verb and resource ('Lists the recordings whose name starts with `prefix`') and then names the exact output content (FPS, 1% low, GPU ms, stutter, hitches, peak Metal MiB) plus the first-to-last delta. It also frames the intended shape of the series ('one recording per version of a translation layer'), which separates it from one-off tools like perf_compare or perf_baseline that it explicitly references.

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 a clear usage context — following a number across a series of runs — and an interpretive rule ('Read a trend as real only when it exceeds the spread_pct that perf_baseline reports for equal runs'), which is real when-to-use guidance. What it lacks is an explicit exclusion: it never says when an agent should choose perf_compare or perf_ab instead of this tool, only that skip_s behaves 'as perf_compare does'.

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