Skip to content

Agent Checks

Agent checks review an agent configuration and surface advisory findings while you build: structural problems (duplicated instructions, conflicting style guidance), completeness gaps (tool references that do not exist), and cost warnings (oversized prompts).

Checks are advisory only. Findings never block saving, publishing, or version creation.

  • Agent editor → Preview tab: a Checks card lists findings for the current draft, updating as you edit.
  • API: POST /v1/agents/preview returns a findings array alongside the resolved system prompt and tools.
  • MCP / platform commands: the preview_agent command returns the same findings, so agents and automations can review configurations programmatically.

Each finding includes:

FieldDescription
rule_idStable rule identifier, e.g. prompt.duplicate_paragraphs
severitywarning, info, or suggestion — there is no error; checks never block
categorystructure, completeness, effectiveness, safety, or cost
messageHuman-readable explanation
locationThe config field (and byte span, when applicable) the finding points at

Checks run against the resolved configuration — after harness and capability contributions are merged — so they can catch issues that span layers.

RuleSeverityWhat it catches
prompt.emptyinfoAgent has no system prompt of its own
prompt.very_longwarningAuthored prompt over 32 KiB, sent on every model turn
prompt.resolved_very_longinfoFull prompt over 96 KiB after harness/capability contributions
prompt.template_variableswarning{{placeholder}} text that would reach the model literally
prompt.duplicate_paragraphswarningThe same paragraph appears more than once
prompt.restates_contributioninfoPrompt duplicates text already contributed by the harness or a capability
prompt.conflicting_styleinfoAsks for both brevity and detail without stating conditions
tools.unknown_referenceinfoPrompt references a tool that no enabled tool or capability provides
tools.duplicate_nameswarningTwo tools share a name, so the model cannot distinguish them

Later phases add on-demand LLM-powered analysis (contradiction and structure review with proposed fixes), behavioral health checks that run the agent against generated smoke tests, and org-configurable rules.