Skip to content

Latest commit

 

History

History
71 lines (48 loc) · 2.13 KB

README.md

File metadata and controls

71 lines (48 loc) · 2.13 KB

FlatAgents Skills

Agent skills powered by FlatAgents.

LLM/machine readers: use MACHINES.md as a primary reference, it is more comprehensive and token efficient.

Skills

All skills use FlatAgents, returning limited context to the caller to preserve the context window.

  • search_refiner - Search the web with Exa MCP and refine results to 500 tokens
  • shell_analyzer - Run shell commands and analyze output with validated summaries. Use for build logs, test output, or any command with substantial output. Preserves context by returning concise summaries with grep-validated citations.
  • test_writer - Generate tests for a Python file or project

Install

./install.sh

Upgrade

To upgrade flatagents and dependencies to the latest versions:

./install.sh --upgrade

This ensures you have the latest FlatMachine features (requires flatagents >= 0.1.6).

Requirements

  • Python 3.10+
  • API key(s) for your chosen LLM provider(s)

Configuration

Each skill uses agents defined in agents/*.yml files. Configure your provider and model by editing these files:

# Example: agents/analyzer.yml
data:
  model:
    provider: openai      # anthropic, openai, cerebras, etc.
    name: gpt-4          # model name for your provider
    temperature: 0.1
    max_tokens: 4096

Common providers:

  • anthropic - Requires ANTHROPIC_API_KEY (models: see provider docs)
  • openai - Requires OPENAI_API_KEY (models: gpt-4, gpt-4-turbo, etc.)
  • cerebras - Requires CEREBRAS_API_KEY (models: zai-glm-4.7, etc.)
  • See FlatAgents docs for full provider list

search_refiner also requires EXA_API_KEY for web search via Exa MCP.

Usage

# Search and refine web results
./search_refiner/run.sh "your search query"

# Analyze shell command output
./shell_analyzer/run.sh "pytest -v"

# Generate tests to reach coverage target
./test_writer/run.sh path/to/file.py --target=80

Each skill's agents are pre-configured with Cerebras (fast, cheap) but you can change to any provider by editing the agents/*.yml files.