Field note
Field notes on repo readiness, execution drift, and agent-safe operations.
These notes stay close to real repository failures: onboarding drift, execution mismatches, and the practical rules that keep one contract usable across teams and agents.
Why AI Agent PRs Get Rejected And How Repo Contracts Help
Rejected AI agent pull requests are often a repo-governance problem, not just a model-quality problem. Repos need explicit setup, verification, workflow, and safety truth.
AI Coding's Real Bottleneck Is Repository Execution Trust
For many AI coding agents, the harder problem is no longer generating code. It is whether a repository exposes a trustworthy setup, verification, and evidence path humans, CI, and agents can all follow.
Why Agent Safety Needs Enforced Boundaries, Not Just Declared Ones
Agent safety does not come from declared safe tasks and protected paths alone. It comes from runners, CI, and runtime boundaries actually enforcing the same repo contract so drift breaks visibly instead of becoming one more soft signal.
Setup Automation Is Not Readiness Verification
Automating setup does not prove a repository is ready. Ota takes a stricter position: repo readiness needs contract-owned verification, explicit execution truth, and machine-readable evidence for humans, CI, and AI agents.
GitHub Managed Settings vs Ota: Platform Governance vs Execution Governance
GitHub Managed Settings standardizes repository configuration. Ota standardizes how repositories are prepared, verified, and run. Modern engineering teams need both.
Why Coding Agents Need Repo Contracts, Not Bigger Context Windows
Bigger context windows help agents read more, but repo contracts help them act safely by making setup, verification, services, env, and task boundaries explicit.
Ota vs Dagger: Portable Workflows Are Not Repo Execution Governance
Dagger makes workflows portable and programmable. Ota makes repositories diagnosable, governable, and provable before, during, and after execution. Workflow portability is useful, but it does not replace repo readiness, safe task boundaries, or receipts.
AGENTS.md Is Not Enough for Safe AI Agent Execution
AGENTS.md can guide an AI coding agent, but it cannot by itself make execution safe, verifiable, or reviewable. Repos also need declared safe commands, canonical verification paths, and receipts that show what actually ran.