From c3fde611d805428734172c99d5eb04fd901c9ef6 Mon Sep 17 00:00:00 2001 From: "claude[bot]" <41898282+claude[bot]@users.noreply.github.com> Date: Wed, 1 Jul 2026 05:23:02 +0000 Subject: [PATCH] docs(TSP-1328): add argument matching to Tool Usage check documentation Documents the new argument conditions capability for Tool Usage evals checks, including the four match types (exact, contains, regex, not empty) and how to configure them in the eval rule form. Co-Authored-By: Claude Sonnet 4.6 --- build/agents/build-your-agent/evals.mdx | 17 +++++++++++++++++ 1 file changed, 17 insertions(+) diff --git a/build/agents/build-your-agent/evals.mdx b/build/agents/build-your-agent/evals.mdx index b7068e12..bef3f47d 100644 --- a/build/agents/build-your-agent/evals.mdx +++ b/build/agents/build-your-agent/evals.mdx @@ -89,6 +89,23 @@ When creating a Check, you choose one of the following types: | **Tool** | Select the tool to check for | | **Position** | Whether the tool was used anywhere, used first, or used last | | **Comparison** | Check if the tool was used at least, exactly, or at most X times | + + #### Argument conditions + + You can add conditions to verify the specific arguments passed to tool calls, not just whether the tool was called. Each condition specifies an argument path and a match type, and all conditions must be satisfied for the check to pass. + + | Match type | Description | + |------------|-------------| + | **Exact match** | The argument value must equal the specified value. Optionally enable case-insensitive matching to ignore capitalization. | + | **Contains** | The argument value must include the specified substring. | + | **Regex** | The argument value must match the specified regular expression pattern. | + | **Not empty** | The argument must have any non-empty value. | + + To add conditions, click **+ Add condition** in the **Conditions** section of the eval rule form. For each condition, specify the argument path and the value or pattern to match against. A plain-English preview of your conditions appears in the form. + + + Argument conditions work for both single-agent evals and workforce (node-scoped) evals. This feature is currently behind the `evals-toolusage-argument-matcher` feature flag. +