This is a template for the Identity Shaping Framework (isf), an agent-first tool for developing AI identities through prompting, evaluation, and fine-tuning.
What is this? A structured template for creating a custom AI identity. You define who the model is, generate training data, and fine-tune a model to embody that identity.
The isf CLI is the main interface. It's command-line driven with extensive help:
isf --help # Top-level commands
isf registry --help # Registry subcommands
isf pipeline --help # Pipeline subcommandsDirectory structure matters. The isf tool expects specific directories and conventions. This template provides that structure - stick to it.
| Resource | Description |
|---|---|
| docs/setup.md | Start here - project setup, configuration |
| docs/workflow.md | Development phases and workflow |
| docs/ | Full documentation |
| Cubs Superfan Example | Fully developed reference template |
| Identity Shaping Framework | The isf tool itself |
identity/ # WHO the model is
SEED.md # Starting point - core concept
IDENTITY.md # Behavioral specification
NARRATIVE.md # Self-narrative and history
templates/ # Jinja2 templates for system prompts
versions/ # Generated prompt versions
pipelines/ # Training data generation
evals/ # Evaluation configs
training/ # Training configs and outputs
results/ # Evaluation results
docs/ # Workflow documentation
isf status # Project overview
isf registry build # Build prompts from identity docs
isf registry release v0.1 # Freeze a version
isf pipeline status # Check pipeline staleness
isf pipeline run NAME # Run a pipeline
isf eval run NAME MODEL # Run evaluation
isf train data prep default # Prepare training data
isf train run CONFIG # Start training