I built this to solve a specific problem: when you run multiple Copilot CLI agents in parallel, you have no way to verify whether their output actually works before it hits your codebase.
Each agent runs on its own git branch. The orchestrator parses Copilot /share transcripts and cross-references every claim (commits, test results, build output) against actual evidence. Steps that can't prove their work don't merge. Failed steps get classified by failure type and retried with targeted repair strategies.
Six quality gates run after merge: scaffold leftovers, duplicate blocks, hardcoded config, README claim drift, test isolation, runtime correctness.
The cost estimator predicts premium request consumption before execution starts, factoring in model multipliers (o3 is 20x, o4-mini is 5x) and historical failure rates. You can preview the estimate or set a hard budget that aborts if exceeded.
v3.2 (just released) replaced the wave-barrier scheduler with greedy as-soon-as-ready scheduling, added prompt compression (~60% token reduction by extracting shared boilerplate into .copilot-instructions.md), and octopus merge for parallel branch completion.
TypeScript, strict mode, 649 tests, no external runtime dependencies beyond Node.js built-ins and the existing dep set. Solo project.
I built this to solve a specific problem: when you run multiple Copilot CLI agents in parallel, you have no way to verify whether their output actually works before it hits your codebase.
Each agent runs on its own git branch. The orchestrator parses Copilot /share transcripts and cross-references every claim (commits, test results, build output) against actual evidence. Steps that can't prove their work don't merge. Failed steps get classified by failure type and retried with targeted repair strategies.
Six quality gates run after merge: scaffold leftovers, duplicate blocks, hardcoded config, README claim drift, test isolation, runtime correctness.
The cost estimator predicts premium request consumption before execution starts, factoring in model multipliers (o3 is 20x, o4-mini is 5x) and historical failure rates. You can preview the estimate or set a hard budget that aborts if exceeded.
v3.2 (just released) replaced the wave-barrier scheduler with greedy as-soon-as-ready scheduling, added prompt compression (~60% token reduction by extracting shared boilerplate into .copilot-instructions.md), and octopus merge for parallel branch completion.
TypeScript, strict mode, 649 tests, no external runtime dependencies beyond Node.js built-ins and the existing dep set. Solo project.