Practical patterns, starters & CLI tools for loop engineering with AI coding agents. Design systems that prompt and orchestrate agents (inspired by Addy Osmani and Boris Cherny). Includes loop-audit, loop-init, loop-cost.
TypeScript agentic-aiai-agentsai-codinganthropicautomationclaudeclaude-codecodexcoding-agentsdevops-automationdevtoolsgithub-actionsgrokllmloop-engineeringmcpprompt-engineering
Decisions and current work
A second clone reads this on its first question — it is one of the decision records knos looks for. Nothing here is private: secrets and private paths never reach it.
Decisions
- shared state without schema — Three loops appending to one unstructured STATE.md causes state rot, conflicting actions and ghost items. One state file per pattern, or clearly separated sections with prune rules. (docs/anti-patterns.md)
- attempt cap — No loop runs "keep trying until CI is green". Hard cap at three attempts, then escalate with full context in the state file. (docs/anti-patterns.md)
- separate verifier — The agent that implements does not verify its own work. A separate verifier sub-agent or model checks it, and the verifier's default stance is REJECT. (docs/anti-patterns.md)
- L1 before L3 — Report-only for the first week. Measure triage accuracy before enabling anything that writes. (docs/anti-patterns.md)
- propose, never merge — Any loop that changes external state opens a draft PR or a comment for a human gate. It never merges. (examples/mcp/safe-write-pattern.md)
- minimum privilege — Loops get the least privilege that works. Prefer read + comment over write; use human gates and worktrees for anything that mutates state. (examples/mcp/README.md)
Being worked on right now
Nothing claimed.
knos export. Claims lapse after 30 minutes.