Dispatch one tool through guardrails — guard

Scans arguments, executes only an allowed request, then scans the tool result before returning it. This is the provider-neutral dispatcher for chat SDKs that do not expose request/result hooks.

Usage

guard_tool(
  tool_name,
  arguments = list(),
  dispatcher,
  tool_policy,
  subject = NULL,
  policy = "enterprise_default",
  reviewer = NULL,
  checks = "rules",
  redaction = NULL,
  scanners = scanner_options(),
  show_tokens = FALSE,
  show_stats = FALSE
)

Arguments

tool_name: Tool name.
arguments: Named argument list.
dispatcher: Function accepting (tool_name, arguments), or a named list of tool functions called with do.call().
tool_policy: Policy from tool_policy().
subject: Optional authorization context passed through unchanged.
policy: A shieldr_policy or built-in policy name.
reviewer: Optional reviewer function or object with $chat().
checks: One of "rules", "nlp", "llm", or "both".
redaction: Optional redaction strategy from redaction_strategy().
scanners: Optional scanner configuration from scanner_options().
show_tokens: Whether to attach token counts when ellmer is available.
show_stats: Show execution statistics as messages.

Value

A list with action, value, call_report, and output_report.

Examples

dispatcher <- list(search = function(query) paste("result", query))
guard_tool(
  "search", list(query = "public"), dispatcher,
  tool_policy(allowed_tools = "search")
)
#> $action
#> [1] "allow"
#> 
#> $value
#> [1] "result public"
#> 
#> $call_report
#> llmshieldr report
#> action: allow
#> risk_score: 0.000
#> findings: 0
#> 
#> $output_report
#> llmshieldr report
#> action: allow
#> risk_score: 0.000
#> findings: 0
#>