How shashuec/conversai-coding-agent-challenge works
This repository presents a coding-agent challenge: the product is an agent that uses a supplied model and constrained repository tools to solve issue-style tasks. Its intended workflow is to inspect and focus source context, request and validate edits, run tests, use failures for repair, then stop. The README documents the interfaces and runner behavior; the agent and runtime source were not sampled, so internal call wiring is limited to those documented contracts.
Main entry points: agent.py Solve loop · runtime.py Repository tools · run_practice.py Challenge runner · practice-repo Task repository
Components 10
Agent orchestration
- Solve loop
agent.py - Context selection
agent.py - Edit validation
agent.py - Retry and stopping
agent.py
Supplied interfaces
- Repository tools
runtime.py - Task repository
practice-repo - Test execution
runtime.py
Task evaluation
- Challenge runner
run_practice.py
Other components
- Task author
- Fixed model
Connections 14
- Task author → Challenge runner: submits task
- Challenge runner → Solve loop: calls solve
- Solve loop → Context selection: selects context
- Context selection → Repository tools: inspects files
- Repository tools → Task repository: reads files
- Solve loop → Fixed model: asks model
- Fixed model → Solve loop: returns proposal
- Solve loop → Edit validation: validates edits
- Edit validation → Repository tools: writes patches
- Repository tools → Task repository: changes source
- Solve loop → Repository tools: runs tests
- Repository tools → Test execution: executes tests
- Test execution → Solve loop: returns failures
- Solve loop → Retry and stopping: checks progress