How tejashiv/agentic-test-repair-system works
This is an agentic test-repair tool: a developer points it at a project, the tool selects failing tests and their source files, asks a model for a repair, applies it, and reruns tests until they pass or attempts run out. The main entry point also supports explicit file pairs, dry runs, and reporting; a separate script runs the QuixBugs benchmark. The excerpts show the core calls and reporting/template wiring. Discovery details and some report-generation links rely on README documentation; unsampled implementation is not inferred beyond that evidence.
Main entry points: agent.py Repair agent · repair.py Repair entry point · quixbugs_runner.py QuixBugs benchmark · runner.py Test runner
Components 13
Repair workflow
- Repair entry point
repair.py - Test discovery and pairing
repair.py
Repair engine
- Repair agent
agent.py - Repair instructions
prompt.txt - Test runner
runner.py
Reports
- Report generation
repair.py - Dashboard template
dashboard_template.py
Benchmark
- QuixBugs benchmark
quixbugs_runner.py - Benchmark summary
quixbugs_runner.py
Other components
- Developer
- Target project
- pytest
- Gemini model
Connections 17
- Developer → Repair entry point: starts repair
README.md - Repair entry point → Test discovery and pairing: discovers or accepts
README.md - Test discovery and pairing → Target project: selects files
README.md - Repair entry point → Test runner: runs tests
repair.py - Repair entry point → Repair agent: repairs failures
repair.py - Repair agent → Test runner: runs tests
agent.py - Repair agent → Repair instructions: loads instructions
agent.py - Repair agent → Target project: reads and writes source
agent.py - Repair agent → Gemini model: requests repair
agent.py - Test runner → pytest: invokes
runner.py - Test runner → Target project: tests in directory
runner.py - Repair entry point → Report generation: builds summary
repair.py - Report generation → Dashboard template: uses template
repair.py - Developer → QuixBugs benchmark: starts benchmark
README.md - QuixBugs benchmark → Repair agent: repairs benchmark cases
quixbugs_runner.py - QuixBugs benchmark → Benchmark summary: summarizes outcomes
quixbugs_runner.py - QuixBugs benchmark → Dashboard template: uses template
quixbugs_runner.py