Internal agent loop turns a board ticket into a reviewed PR
alex_verem · x · 2026-07-22
This post describes an internal “Velocity Core” workflow that turns a board ticket into a reviewed pull request with no developer writing code directly.
- A developer marks a ticket as AI: Ready and the orchestrator gathers repo and code-graph context.
- Each task runs in its own disposable sandbox with scoped credentials.
- The agent plans, edits, runs tests, reads CI feedback, and iterates until tests pass or a ceiling is reached and a human is flagged.
- The goal is an end-to-end loop from task intake to reviewed PR, with automation plus human escalation when needed.
More from coding & agent
- Peer review should not receive AI-generated vibe reports, the thread argues — fredahshi · 2026-07-22
- AI Dev Tools Zoomcamp opens with a workshop on spec, context and loop engineering — Al_Grigor · 2026-07-22
- freeCodeCamp shows how to test flaky AI agents with repeatable Python evals — solyarisoftware · 2026-07-22
- Prompt library idea: delete rules, test variants, keep only what matters — _ScottCondron · 2026-07-22
- LLM-as-a-Verifier uses continuous scores to improve agent evaluation — solyarisoftware · 2026-07-22
- Open-source `CLAUDE.md` file turns Karpathy’s coding advice into guardrails for Claude Code — socialwithaayan · 2026-07-22