Skip to contents

Install codeagent's central permission gate on an existing Chat

Source: R/tools_gate.R
install_permission_gate.Rd

For hosts that build a harness-only client (codeagent_client(register_tools = FALSE)) and attach their own domain tools, then want those tools governed by codeagent's central permission gate.

Declare each sensitive tool's capability with register_tool_meta() (or pass tool_meta here), then call this once after attaching the tools. Tools whose capability resolves to "read" are allowed without gating; write/exec/ net route through the gate under permission_mode + the tools policy.

Idempotent per chat: calling again refreshes the live mode / ask_fn / policy rather than stacking a second gate.

Usage

install_permission_gate(
  chat,
  permission_mode = "default",
  rules = list(),
  tools = list(),
  ask_fn = NULL,
  tool_meta = list()
)

Arguments

chat

An ellmer::Chat.

permission_mode

Character. One of the 7 modes (default "default").

rules

List. Fine-grained permission rules (see check_permission()).

tools

List. settings$tools policy (sets / capabilities / overrides).

ask_fn

Function or NULL. Permission callback invoked when a tool needs approval; may return a logical or a promise (async / Shiny). Called as ask_fn(name, input), or ask_fn(name, input, id = <tool_call_id>) if it declares an id argument or ....

tool_meta

Named list mapping tool name -> capability ("read"/"write"/"exec"/"net"), a convenience for calling register_tool_meta() on each before installing the gate.

Value

Invisibly, chat.

See also

register_tool_meta(), codeagent_client()

Examples

if (FALSE) { # \dontrun{
client <- codeagent_client(chat, register_tools = FALSE)
chat$register_tool(my_write_tool)
install_permission_gate(
  chat, permission_mode = "default",
  tool_meta = list(MyWriteTool = "write"),
  ask_fn = function(name, input, id = NULL) host_prompt(id, name, input))
} # }