Overview

• module Brute
  • class Agent

    An agent is a Pipeline configured for LLM turns. It carries the
  • module Events
    • class Handler

      Stackable event handler base class. Subclasses override the
    • class PrefixedTerminalOutput

      TerminalOutput variant that prefixes all output with a label.
    • class TerminalOutput
  • module Middleware
    • class OtelSpan

      Outermost OTel middleware. Creates a span per LLM stack call
    • class ToolResultLoop

      Re-invokes the inner stack whenever the last message is a :tool result.
    • class Summarize

      Runs a final tool-free LLM call after the ToolResultLoop completes,
    • class Tracing

      Logs timing and token usage for every LLM call, and tracks cumulative
    • class MaxIterations

      Guards against runaway tool loops by capping the number of iterations.
    • module OTel
      • class TokenUsage

        Records token usage from the LLM response as span attributes.
    • class SystemPrompt

      Prepends a system message to env[:messages] before passing control
    • class CompactionCheck

      Checks context size after each LLM call and triggers compaction

      • class Compactor

        Context compaction service. When the conversation grows past configurable
    • class Question
    • class ToolCall

      Executes pending tool calls from the LLM response.
    • class OtelToolCalls

      Records tool calls the LLM requested as span events.
    • class OtelToolResults

      Records tool results as span events.
    • class LLMCall

      Terminal middleware. Calls the LLM with the current conversation,
    • class EventHandler
    • class UserQueue
  • class Pipeline

    Generic middleware machinery. Builds a chain of middleware around

    • class NullSink

      Default null sink for env[:events] — swallows anything pushed to it.
  • module Prompts
    • module Autonomy
    • module BuildSwitch
    • module CodeReferences
    • module CodeStyle
    • module Conventions
    • module DoingTasks
    • module EditingApproach
    • module EditingConstraints
    • module Environment
    • module FrontendTasks
    • module GitSafety
    • module Identity
    • module Instructions
    • module MaxSteps
    • module Objectivity
    • module PlanReminder
    • module Proactiveness
    • module SecurityAndSafety
    • module Skills
    • module TaskManagement
    • module ToneAndStyle
    • module ToolUsage
  • module Providers
    • class Shell

      A pseudo-LLM provider that executes user input as code via the

      • class ModelList

        Minimal object that quacks like provider.models so the REPL's
    • class ShellResponse

      Synthetic completion response returned by Brute::Providers::Shell.
  • module Queue
    • module FileMutationQueue

      Per-file serialization queue for concurrent tool execution.

      • class << self
  • class Session
  • module Skill

    Discovers and loads SKILL.md files from standard directories.
  • module Store
    • module SnapshotStore

      Per-path stack of file snapshots used by fs_write, fs_patch, fs_remove

      • class << self
    • module TodoStore

      In-memory todo list storage. The agent uses this to track multi-step tasks.

      • class << self
  • class SubAgent

    A SubAgent is an Agent that exposes a tool-shaped facade so it can
  • class SystemPrompt

    Deferred system prompt builder.
  • class Tool

    A Tool is a Pipeline configured for tool execution. The terminal app
  • module Tools
    • class FSPatch
    • class FSRead

      Existing features (ref: opencode read tool):
    • class FSRemove
    • class FSSearch

      Existing features (ref: opencode grep tool):
    • class FSUndo
    • class FSWrite
    • class NetFetch
    • class Question
    • class Shell

      Existing features (ref: opencode bash tool):
    • class TodoRead
    • class TodoWrite
  • module Truncation

    Universal tool output truncation.
  • module Diff
• test
  • it
    • mktmpdir
      • each

        Each result line should be capped, not the full 3000 chars
• each