How seratul03/aria works
ARIA is an autonomous agent that runs registered tools, records their outcomes, identifies weaknesses, and attempts safe improvements. Its main improvement path moves from the agent through introspection and code generation to evolutionary evaluation, Gatekeeper validation, and Git deployment; the README documents the candidate arena and referee as well. A separate meta-improvement path models system-wide weaknesses, modifies isolated clones, and compares them in an arena. Memory, knowledge, prediction, root-cause analysis, and reflection support these loops. Edges follow the README or shown source/index references; unsampled implementation details are not inferred.
Main entry points: agent.py Agent Core · meta.py Meta-Improvement · engine.py Weakness Analysis · arena.py Candidate Arena
Components 23
Agent Runtime
- Agent Core
aria/core/agent.py - Priority Scheduler
aria/core/scheduler.py - Tool Registry
aria/tools/registry.py - Improvable Tools
aria/tools - Metrics Store
aria/metrics/db.py - Cycle Tracing
aria/core/tracer.py
Tool Improvement
- Weakness Analysis
aria/introspection/engine.py - Improvement Engine
aria/improvement/engine.py - Candidate Arena
aria/evolution/arena.py - Code Gatekeeper
aria/gatekeeper - Arena Referee
aria/gatekeeper/referee/server.py - Git Deployment
aria/versioning/git_manager.py
Meta Improvement
- Meta-Improvement
aria/introspection/meta.py - Isolated Clones
aria/introspection/clone.py - Self-Model and WTDS
aria/introspection/self_model.py
Learning and Analysis
- Failure Memory
aria/memory/store.py - Engineering Knowledge
aria/knowledge/applications.py - Evolution Predictors
aria/predictors/inference.py - Root-Cause Analysis
aria/rootcause/classifier.py - Reflection Analysis
aria/reflection/weaknesses.py
User Interfaces
- CLI Commands
aria/cli.py - Interactive Menu
aria/ui/interactive_menu.py
Other components
- Operator
Connections 29
- Operator → CLI Commands: runs commands
README.md - Operator → Interactive Menu: uses
README.md - CLI Commands → Agent Core: dispatches tasks
aria/cli.py - Interactive Menu → Agent Core: controls
aria/ui/interactive_menu.py - Agent Core → Tool Registry: looks up tools
aria/core/agent.py - Agent Core → Metrics Store: records executions
aria/core/agent.py - Tool Registry → Improvable Tools: registers
aria/tools/registry.py - Agent Core → Priority Scheduler: selects targets
README.md - Priority Scheduler → Weakness Analysis: finds weaknesses
README.md - Weakness Analysis → Improvement Engine: provides weakness report
aria/improvement/engine.py - Improvement Engine → Candidate Arena: generates candidates
README.md - Candidate Arena → Code Gatekeeper: validates candidates
README.md - Candidate Arena → Arena Referee: submits combat results
README.md - Code Gatekeeper → Git Deployment: deploys accepted code
README.md - Arena Referee → Git Deployment: deploys winner
README.md - Agent Core → Weakness Analysis: starts cycles
aria/core/agent.py - Agent Core → Evolution Predictors: checks viability
aria/core/agent.py - Evolution Predictors → Candidate Arena: ranks candidates
aria/predictors/inference.py - Agent Core → Cycle Tracing: records cycle traces
aria/core/agent.py - Cycle Tracing → Metrics Store: persists traces
aria/core/tracer.py - Weakness Analysis → Failure Memory: retrieves failure context
aria/introspection/engine.py - Agent Core → Engineering Knowledge: applies engineering rules
aria/core/agent.py - Meta-Improvement → Self-Model and WTDS: updates system model
README.md - Weakness Analysis → Self-Model and WTDS: analyzes traces
README.md - Meta-Improvement → Isolated Clones: creates isolated copies
README.md - Isolated Clones → Candidate Arena: competes in arena
README.md - Meta-Improvement → Arena Referee: requests scoring
README.md - Meta-Improvement → Reflection Analysis: uses weakness findings
aria/introspection/meta.py - Meta-Improvement → Root-Cause Analysis: uses root-cause findings
aria/introspection/meta.py