Genesis V2 implements a direct Execute-Refine loop with automatic quality assessment and iterative improvement. It runs a single direct agent with skills and refines based on evaluator feedback.
Usage
genesis_v2(
task,
skill_paths = "auto",
model = "claude-3-5-sonnet-20241022",
max_iterations = 3,
auto_refine = TRUE,
quality_threshold = 70,
cache = TRUE,
verbose = FALSE,
architect_model = NULL,
evaluator_model = NULL,
refiner_model = NULL,
max_steps = 10,
use_computer_tools = FALSE
)Arguments
- task
Character string describing the task to accomplish
- skill_paths
Character vector of paths to scan for skills, or "auto" for default locations
- model
Model to use for agents (default: claude-3-5-sonnet-20241022)
- max_iterations
Maximum number of iterations (default: 3)
- auto_refine
Logical, whether to enable automatic refinement (default: TRUE)
- quality_threshold
Minimum quality score to accept (0-100, default: 70)
- cache
Logical, whether to cache team composition for similar tasks (unused in direct mode)
- verbose
Logical, whether to print orchestration details (default: FALSE)
- architect_model
Model to use for criteria generation (default: same as model)
- evaluator_model
Model to use for Evaluator agent (default: same as model)
- refiner_model
Model to use for Refiner agent (unused in direct mode)
- max_steps
Maximum tool execution steps (default: 10)
- use_computer_tools
Logical, whether to use computer abstraction layer (default: FALSE). When TRUE, uses atomic tools (bash, read_file, write_file, execute_r_code) instead of loading all skill tools into context. This reduces context window usage by 30-50%.
Examples
if (FALSE) { # \dontrun{
result <- genesis_v2(
"Analyze the iris dataset and create a comprehensive report with visualizations",
max_iterations = 3,
quality_threshold = 80,
auto_refine = TRUE,
verbose = TRUE
)
# With computer tools for reduced context usage
result <- genesis_v2(
"Complex multi-step analysis",
use_computer_tools = TRUE,
max_iterations = 5
)
} # }