Compaction, cost tracking, -O optimization and more

[jgarzik]

No description provided.

[Copilot]

Pull request overview

This PR implements multiple major features including token optimization mode (-O flag), comprehensive cost tracking, context compaction, user-defined slash commands, and HTTP/SSE transport support for MCP servers. It also consolidates tool filtering logic and adds configuration validation.

Key changes:

• Token optimization system with compressed tool schemas and AI-to-AI communication mode
• Session-level cost tracking with per-model pricing and turn-by-turn breakdown
• Context compaction to manage conversation history and reduce token usage
• User-defined slash commands loaded from .yo/commands/ directories
• HTTP and SSE transport support for MCP servers alongside existing stdio
• Configuration validation with detailed error reporting
• Consolidated tool pattern matching logic into dedicated module

Reviewed changes

Copilot reviewed 28 out of 29 changed files in this pull request and generated 17 comments.

Show a summary per file

File	Description
src/transcript.rs	Adds token_usage logging method for tracking LLM API calls
src/tools/write.rs	Implements schema optimization with terse descriptions when -O flag is set
src/tools/task.rs	Adds optimized schema support for Task tool delegation
src/tools/read.rs	Implements compressed schema for file reading operations
src/tools/mod.rs	Introduces SchemaOptions struct for controlling schema verbosity
src/tools/grep.rs	Adds schema optimization for grep search operations
src/tools/glob.rs	Implements terse schema for file pattern matching
src/tools/edit.rs	Adds optimized schema for file editing operations
src/tools/bash.rs	Implements schema optimization and adds test for optimized mode
src/tools/activate_skill.rs	Adds schema optimization for skill pack activation
src/tool_filter.rs	New module consolidating tool pattern matching logic from policy and subagent
src/subagent.rs	Integrates tool filtering module, adds cost tracking and optimization mode support
src/policy.rs	Refactors to use consolidated tool_filter module
src/plan.rs	Optimizes regex compilation with once_cell caching
src/mcp/transport.rs	Adds HTTP and SSE transport implementations alongside stdio
src/mcp/manager.rs	Updates to support multiple transport types (stdio/HTTP/SSE)
src/mcp/client.rs	Abstracts transport layer with McpTransportImpl enum
src/main.rs	Adds new modules, command index initialization, and config validation
src/llm.rs	Introduces LlmClient trait for abstraction and testing
src/hooks.rs	Removes trailing whitespace
src/cost.rs	New comprehensive cost tracking system with model pricing and session management
src/config.rs	Adds validation framework, MCP transport config, and cost tracking configuration
src/compact.rs	New context compaction module for conversation history summarization
src/commands.rs	New slash commands system for user-defined markdown command templates
src/cli.rs	Integrates cost display, compaction, slash commands, and turn counter
src/agent.rs	Adds cost tracking, optimization mode instructions, and force_continue support
IDEAS.md	Documents future optimization strategies based on LLM research
Cargo.toml	Adds once_cell dependency for regex caching
Cargo.lock	Updates lock file with once_cell dependency

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.