fableforge-agent-profiler 0.1.0

Profile and classify agent behavior patterns from traces

Agent Profiler

Profile and classify agent behavior patterns from traces using transition matrices and tool distributions.

Installation

pip install agent-profiler

Quick Start

Profile a Session

# Profile and classify behavior
aprof profile trace.jsonl

# Save results to JSON
aprof profile trace.jsonl -o results.json

Classify Behavior

aprof classify trace.jsonl

Generate Visualizations

aprof visualize trace.jsonl --output profile_chart.png

Programming API

from agent_profiler import AgentProfiler, BehaviorClassifier, ProfileVisualizer

# Profile a session
profiler = AgentProfiler()
result = profiler.profile("trace.jsonl")
print(f"Category: {result.category} (confidence: {result.confidence:.1%})")
print(f"Tool distribution: {result.tool_distribution.tool_counts}")

# Classify directly
classifier = BehaviorClassifier()
category, confidence, scores = classifier.classify(
    edit_ratio=0.25, read_ratio=0.15, bash_ratio=0.30,
    error_rate=0.4, error_recovery_rate=0.35
)

# Generate visualizations
visualizer = ProfileVisualizer()
visualizer.generate_profile_chart(result, output="profile.png")
visualizer.generate_transition_heatmap("trace.jsonl", output="heatmap.png")
visualizer.generate_tool_distribution_pie("trace.jsonl", output="tools.png")

Behavior Categories

Category	Description	Key Indicators
Debugging	Active debugging sessions	High Edit+Bash, many errors, recoveries
Building	Feature development	High Write+Bash, low Read
Exploring	Code investigation	High Read+Grep, low Edit
Lost	Confused/circular behavior	Circular transitions, high Read
Verifying	Change verification	Read after Edit, test runs

License

MIT

Ecosystem

Part of the FableForge ecosystem — 21 open-source projects built from 210K real agent traces:

Project	Description
Anvil	Self-verified coding agent
VerifyLoop	Plan→Execute→Verify→Recover framework
ErrorRecovery	Self-healing middleware (3,725 error patterns)
FableForge-14B	The fine-tuned 14B model (4-stage training)
ShellWhisperer	1.5B edge agent (phone/RPi, 50ms)
ReasonCritic	Verification model (130 benchmark tasks)
TraceCompiler	Compile traces → LoRA skills
AgentRuntime	Persistent agent daemon (systemd for AI)
AgentSwarm	Multi-agent from real trace transitions
AgentTelemetry	Datadog for agents (token tracking, costs)
BenchAgent	HumanEval for tool-use (107 tasks)
AgentDev	VSCode extension with verification
TraceViz	Trace replay visualizer (Next.js)
AgentSkills	npm for agent behaviors
AgentCurriculum	5-stage progressive training
AgentFuzzer	Adversarial testing for agents
AgentConstitution	Safety guardrails from traces
CostOptimizer	Token cost reduction (50-80%)
AgentProfiler	Behavioral fingerprinting
TrajectoryDistiller	Trace→training data pipeline
Fable5-Dataset	HuggingFace dataset release