small changes to prompt and caching

improve reflection loop and add cache

iterations to improve defineScope and gatherContext

refine prompts for each phase

use different llms for each phase

refactor tools and rank files in define scope

fix: sandbox script and prompts

fix: use createllm

fix: revert the changes from the auditor agent and update the poc agent

feat: integrate the agents

feat(tester): implement PoC generator with DeepSeek and Foundry loop

fix: change the tests

docs: add readme and change the foundryRunner

tests: add e2e test for the agent

feat: Validated Implementation with Centrifuge Case

feat: Implementar `scaffoldGenerator` and Implementar `oracleNode`

feat: Criar nodes stub e grafo linear

implement base agents

wip: auditor agent

setup project