optimizer_skill

Overview

This skill analyzes an existing agent codebase and automatically applies targeted patches to improve performance, quality, security, and functionality. It prioritizes the agent's system_prompt and mcp_config.py to correct behavior and expand capabilities, using AST-driven analysis and atomic code edits. The workflow includes discovery, deep analysis, snapshotting, patching, verification, and re-registration of the optimized agent.

How this skill works

The skill locates the target agent via the agent registry, then runs AST-based inspections to surface structural, complexity, performance, and security issues. It crafts focused, single-file patch operations and uses a snapshot-before-change strategy to ensure safe rollbacks. After applying patches, it verifies results and dynamically re-registers the updated agent for immediate use.

When to use it

• Agent produces incorrect or inconsistent behavior (wrong steps, missed constraints)
• Agent lacks a capability due to missing tools in mcp_config.py
• Code has clear performance or security hotspots discovered by static analysis
• You need a safe, auditable way to change an agent (snapshots + atomic patches)
• You want to fuse patterns from built-in agents to create a specialized agent

Best practices

• Always prioritize fixes to system_prompt and mcp_config.py before changing runtime logic
• Use regular-expression queries for all AST searches to get accurate locations and context
• Create a compressed snapshot of the agent directory before any modifications
• Make each patch atomic: one logical block in one file per operation
• Include the terminal tool in mcp_config.py for dependency management and file I/O operations

Example use cases

• Fix an agent that skips required workflow steps by strengthening the Methodology & Workflow section in system_prompt
• Add a web-search or PDF-reading tool to mcp_config.py to enable new capabilities
• Reduce a function's high cyclomatic complexity by refactoring a single function and applying an atomic patch
• Inject a dynamic current-date variable into system_prompt so the agent is aware of the correct time zone and cutoff date
• Fuse the foundation search agent prompt with a specialist agent's domain logic to create a tuned, high-quality agent